Chapter Fifteen

OPTIMIZATION: Contents
15.1 Critical Points: Local Extrema and
LOCAL AND Saddle Points . . . . . . . . . . . . . . . . . . . . . . . . . . . .
How Do We Detect a Local Maximum
806

GLOBAL or Minimum? . . . . . . . . . . . . . . . . . . . . . . . . . . .
Finding and Analyzing Critical Points . . . . . . . . . .
806
807
Classifying Critical Points . . . . . . . . . . . . . . . . . . 809
EXTREMA Quadratic Functions of the Form
f (x, y) = ax2 + bxy + cy2 . . . . . . . . . . . . . . . . . . . 809
The Shape of the Graph of
f (x, y) = ax2 + bxy + cy2 . . . . . . . . . . . . . . . . . . . 810
Classifying the Critical Points of a Function . . . . . 811
15.2 Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 815
How Do We Find Global Maxima and Minima? . 816
Maximizing Profit and Minimizing Cost . . . . . . 816
Fitting a Line to Data: Least Squares . . . . . . . . . 819
How Do We Know Whether a Function
Has a Global Maximum or Minimum? . . . . . . . 820
15.3 Constrained Optimization:
Lagrange Multipliers . . . . . . . . . . . . . . . . . . . . . 825
Graphical Approach: Maximizing Production
Subject to a Budget Constraint . . . . . . . . . . . . . 825
Analytical Solution: Lagrange Multipliers . . . . . 826
Lagrange Multipliers in General . . . . . . . . . . . . . 827
How to Distinguish Maxima from Minima . . . . . . . 828
Optimization with Inequality Constraints . . . . . . 828
The Meaning of λ . . . . . . . . . . . . . . . . . . . . . . . . 830
An Expression for λ . . . . . . . . . . . . . . . . . . . . . . 830
The Lagrangian Function. . . . . . . . . . . . . . . . . . . 831
806 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

15.1 CRITICAL POINTS: LOCAL EXTREMA AND SADDLE POINTS

Functions of several variables, like functions of one variable, can have local and global extrema.
(That is, local and global maxima and minima.) A function has a local extremum at a point where
it takes on the largest or smallest value in a small region around the point. Global extrema are the
largest or smallest values anywhere on the domain under consideration. (See Figures 15.1 and 15.2.)
𝑦
𝑧 𝑏

11
0 6
3
1
𝑦 1

−3

𝑥 −1 𝑥
𝑎

Figure 15.1: Local and global extrema for a Figure 15.2: Contour map of the function
function of two variables on 0 ≤ 𝑥 ≤ 𝑎, in Figure 15.1
0≤𝑦≤𝑏

More precisely, considering only points at which 𝑓 is defined, we say:

• 𝑓 has a local maximum at the point 𝑃0 if 𝑓 (𝑃0 ) ≥ 𝑓 (𝑃 ) for all points 𝑃 near 𝑃0 .
• 𝑓 has a local minimum at the point 𝑃0 if 𝑓 (𝑃0 ) ≤ 𝑓 (𝑃 ) for all points 𝑃 near 𝑃0 .

For example, the function whose contour map is shown in Figure 15.2 has a local minimum
value of −3 and local maximum values of 3 and 11 in the rectangle shown.

How Do We Detect a Local Maximum or Minimum?

Recall that if the gradient vector of a function is defined and nonzero, then it points in a direction in
which the function increases. Suppose that a function 𝑓 has a local maximum at a point 𝑃0 which
is not on the boundary of the domain. If the vector grad 𝑓 (𝑃0 ) were defined and nonzero, then we
could increase 𝑓 by moving in the direction of grad 𝑓 (𝑃0 ). Since 𝑓 has a local maximum at 𝑃0 , there
is no direction in which 𝑓 is increasing. Thus, if grad 𝑓 (𝑃0 ) is defined, we must have

grad 𝑓 (𝑃0 ) = 0⃗ .
Similarly, suppose 𝑓 has a local minimum at the point 𝑃0 . If grad 𝑓 (𝑃0 ) were defined and nonzero,
then we could decrease 𝑓 by moving in the direction opposite to grad 𝑓 (𝑃0 ), and so we must again
have grad 𝑓 (𝑃0 ) = 0⃗ . Therefore, we make the following definition:

Points where the gradient is either 0⃗ or undefined are called critical points of the function.

If a function has a local maximum or minimum at a point 𝑃0 , not on the boundary of its domain,
then 𝑃0 is a critical point. For a function of two variables, we can also see that the gradient vector must
be zero or undefined at a local maximum by looking at its contour diagram and a plot of its gradient
vectors. (See Figures 15.3 and 15.4.) Around the maximum the vectors are all pointing inward,
perpendicularly to the contours. At the maximum the gradient vector must be zero or undefined. A
similar argument shows that the gradient must be zero or undefined at a local minimum.
 15.1 CRITICAL POINTS: LOCAL EXTREMA AND SADDLE POINTS 807

𝑦 𝑦

5
𝑥 𝑥

Figure 15.3: Contour diagram around a Figure 15.4: Gradients pointing toward the local maximum of
local maximum of a function the function in Figure 15.3
Finding and Analyzing Critical Points
⃗ = 0⃗ , which means setting all the
To find critical points of 𝑓 we set grad 𝑓 = 𝑓𝑥 ⃗𝑖 + 𝑓𝑦 𝑗⃗ + 𝑓𝑧 𝑘
partial derivatives of 𝑓 equal to zero. We must also look for the points where one or more of the
partial derivatives is undefined.

Example 1 Find and analyze the critical points of 𝑓 (𝑥, 𝑦) = 𝑥2 − 2𝑥 + 𝑦2 − 4𝑦 + 5.

Solution To find the critical points, we set both partial derivatives equal to zero:

𝑓𝑥 (𝑥, 𝑦) = 2𝑥 − 2 = 0
𝑓𝑦 (𝑥, 𝑦) = 2𝑦 − 4 = 0.

Solving these equations gives 𝑥 = 1, 𝑦 = 2. Hence, 𝑓 has only one critical point, namely (1, 2). To
see the behavior of 𝑓 near (1, 2), look at the values of the function in Table 15.1.
Table 15.1 Values of 𝑓 (𝑥, 𝑦) near the point (1, 2)
𝑥
0.8 0.9 1.0 1.1 1.2
1.8 0.08 0.05 0.04 0.05 0.08
1.9 0.05 0.02 0.01 0.02 0.05
𝑦 2.0 0.04 0.01 0.00 0.01 0.04
2.1 0.05 0.02 0.01 0.02 0.05
2.2 0.08 0.05 0.04 0.05 0.08

The table suggests that the function has a local minimum value of 0 at (1, 2). We can confirm
this by completing the square:

𝑓 (𝑥, 𝑦) = 𝑥2 − 2𝑥 + 𝑦2 − 4𝑦 + 5 = (𝑥 − 1)2 + (𝑦 − 2)2 .

Figure 15.5 shows that the graph of 𝑓 is a paraboloid with vertex at the point (1, 2, 0). It is the
same shape as the graph of 𝑧 = 𝑥2 + 𝑦2 (see Figure 12.12 on page 661), except that the vertex has
been shifted to (1, 2). So the point (1, 2) is a local minimum of 𝑓 (as well as a global minimum).
𝑧

𝑥
𝑦
(1, 2, 0)

Figure 15.5: The graph of 𝑓 (𝑥, 𝑦) = 𝑥2 − 2𝑥 + 𝑦2 − 4𝑦 + 5 with a local minimum at the point (1, 2)
808 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

√
Example 2 Find and analyze any critical points of 𝑓 (𝑥, 𝑦) = − 𝑥2 + 𝑦2 .

Solution We look for points where grad 𝑓 = 0⃗ or is undefined. The partial derivatives are given by
𝑥
𝑓𝑥 (𝑥, 𝑦) = − √ ,
𝑥2 + 𝑦2
𝑦
𝑓𝑦 (𝑥, 𝑦) = − √ .
𝑥2 + 𝑦2

These partial derivatives are never simultaneously zero, but they are undefined at 𝑥 = 0, 𝑦 = 0.
Thus, (0, 0) is a critical point and a possible extreme point. The graph of 𝑓 (see Figure 15.6) is a
cone, with vertex at (0, 0). So 𝑓 has a local and global maximum at (0, 0).
𝑧

𝑥 ✯ 𝑦
Local maximum
Global maximum

√
Figure 15.6: Graph of 𝑓 (𝑥, 𝑦) = − 𝑥2 + 𝑦2

Example 3 Find and analyze any critical points of 𝑔(𝑥, 𝑦) = 𝑥2 − 𝑦2 .

Solution To find the critical points, we look for points where both partial derivatives are zero:

𝑔𝑥 (𝑥, 𝑦) = 2𝑥 = 0
𝑔𝑦 (𝑥, 𝑦) = −2𝑦 = 0.

Solving gives 𝑥 = 0, 𝑦 = 0, so the origin is the only critical point.

Figure 15.7 shows that near the origin 𝑔 takes on both positive and negative values. Since
𝑔(0, 0) = 0, the origin is a critical point which is neither a local maximum nor a local minimum. The
graph of 𝑔 looks like a saddle.
𝑧
𝑧

𝑥 𝑦
Figure 15.7: Graph of
𝑔(𝑥, 𝑦) = 𝑥2 − 𝑦2 , showing saddle Figure 15.8: Graph of ℎ(𝑥, 𝑦) = 𝑥2 + 𝑦2 , showing minimum
shape at the origin at the origin
The previous examples show that critical points can occur at local maxima or minima, or at
points which are neither: The functions 𝑔 and ℎ in Figures 15.7 and 15.8 both have critical points
at the origin. Figure 15.9 shows level curves of 𝑔. They are hyperbolas showing both positive and
negative values of 𝑔 near (0, 0). Contrast this with the level curves of ℎ near the local minimum in
Figure 15.10.
 15.1 CRITICAL POINTS: LOCAL EXTREMA AND SADDLE POINTS 809
𝑦 𝑦
3 3
−4 12
2 2 8
−2 4

1 1

4
2
𝑥 𝑥
2
0
4
0
−1 −1 2
−2
−2 −4 −2 6
10
−3 −3
−3 −2 −1 1 2 3 −3 −2 −1 1 2 3
Figure 15.9: Contours of 𝑔(𝑥, 𝑦) = 𝑥2 − 𝑦2 , Figure 15.10: Contours of ℎ(𝑥, 𝑦) = 𝑥2 + 𝑦2 ,
showing a saddle shape at the origin showing a local minimum at the origin

Example 4 Find the local extrema of the function 𝑓 (𝑥, 𝑦) = 8𝑦3 + 12𝑥2 − 24𝑥𝑦.

Solution We begin by looking for critical points:

𝑓𝑥 (𝑥, 𝑦) = 24𝑥 − 24𝑦,

𝑓𝑦 (𝑥, 𝑦) = 24𝑦2 − 24𝑥.

Setting these expressions equal to zero gives the system of equations

𝑥 = 𝑦, 𝑥 = 𝑦2 ,

which has two solutions, (0, 0) and (1, 1). Are these local maxima, local minima or neither? Fig-
ure 15.11 shows contours of 𝑓 near the points. Notice that 𝑓 (1, 1) = −4 and the contours at nearby
points have larger function values. This suggests 𝑓 has a local minimum at (1, 1).
We have 𝑓 (0, 0) = 0 and the contours near (0, 0) show that 𝑓 takes both positive and negative
values nearby. This suggests that (0, 0) is a critical point which is neither a local maximum nor a
local minimum.
𝑦
2

1.5
−2

1
−4
−3
0.5
3

−1
1
0
2 𝑥
0

−0.5
−0.5 0.5 1 1.5 2

Figure 15.11: Contour diagram of 𝑓 (𝑥, 𝑦) = 8𝑦3 + 12𝑥2 − 24𝑥𝑦 showing critical points at (0, 0) and (1, 1)

Classifying Critical Points

We can see whether a critical point of a function, 𝑓 , is a local maximum, local minimum, or neither
by looking at the contour diagram. There is also an analytic method for making this distinction.
Quadratic Functions of the Form 𝒇 (𝒙, 𝒚) = 𝒂𝒙𝟐 + 𝒃𝒙𝒚 + 𝒄𝒚 𝟐
Near most critical points, a function has the same behavior as its quadratic Taylor approximation
810 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

about that point. Thus, we start by investigating critical points of quadratic functions of the form
𝑓 (𝑥, 𝑦) = 𝑎𝑥2 + 𝑏𝑥𝑦 + 𝑐𝑦2 , where 𝑎, 𝑏 and 𝑐 are constants.

Example 5 Find and analyze the local extrema of the function 𝑓 (𝑥, 𝑦) = 𝑥2 + 𝑥𝑦 + 𝑦2 .

Solution To find critical points, we set

𝑓𝑥 (𝑥, 𝑦) = 2𝑥 + 𝑦 = 0,
𝑓𝑦 (𝑥, 𝑦) = 𝑥 + 2𝑦 = 0.

The only critical point is (0, 0), and the value of the function there is 𝑓 (0, 0) = 0. If 𝑓 is always
positive or zero near (0, 0), then (0, 0) is a local minimum; if 𝑓 is always negative or zero near
(0, 0), it is a local maximum; if 𝑓 takes both positive and negative values, it is neither. The graph in
Figure 15.12 suggests that (0, 0) is a local minimum.
How can we be sure that (0, 0) is a local minimum? We complete the square. Writing

1 2 3
( )
𝑓 (𝑥, 𝑦) = 𝑥2 + 𝑥𝑦 + 𝑦2 = 𝑥 + 𝑦 + 𝑦2 ,
2 4
shows that 𝑓 (𝑥, 𝑦) is a sum of two nonnegative terms, so it is always greater than or equal to zero.
Thus, the critical point is both a local and a global minimum.
𝑧

𝑦
✒ 𝑥
Local minimum

Figure 15.12: Graph of 𝑓 (𝑥, 𝑦) = 𝑥2 + 𝑥𝑦 + 𝑦2 = (𝑥 + 21 𝑦)2 + 34 𝑦2 showing local minimum at the origin

The Shape of the Graph of 𝒇 (𝒙, 𝒚) = 𝒂𝒙𝟐 + 𝒃𝒙𝒚 + 𝒄𝒚 𝟐

In general, a function of the form 𝑓 (𝑥, 𝑦) = 𝑎𝑥2 + 𝑏𝑥𝑦 + 𝑐𝑦2 has one critical point at (0, 0). Assuming
𝑎 ≠ 0, we complete the square and write
𝑏 2
[( ( ) ]
𝑏 𝑐 2 𝑐 𝑏2
[ ] )
2 2 2
𝑎𝑥 + 𝑏𝑥𝑦 + 𝑐𝑦 = 𝑎 𝑥 + 𝑥𝑦 + 𝑦 = 𝑎 𝑥 + 𝑦 + − 𝑦2
𝑎 𝑎 2𝑎 𝑎 4𝑎2
𝑏 2
[( ( ) ]
4𝑎𝑐 − 𝑏2
)
=𝑎 𝑥+ 𝑦 + 𝑦2 .
2𝑎 4𝑎2
The shape of the graph of 𝑓 depends on whether the coefficient of 𝑦2 is positive, negative, or zero.
The sign of the discriminant, 𝐷 = 4𝑎𝑐 − 𝑏2 , determines the sign of the coefficient of 𝑦2 .
• If 𝐷 > 0, then the expression inside the square brackets is positive or zero, so the function has
a local maximum or a local minimum.
∙ If 𝑎 > 0, the function has a local minimum, since the graph is a paraboloid opening upward,
like 𝑧 = 𝑥2 + 𝑦2 . (See Figure 15.13.)
∙ If 𝑎 < 0, the function has a local maximum, since the graph is a paraboloid opening down-
ward, like 𝑧 = −𝑥2 − 𝑦2 . (See Figure 15.14.)
• If 𝐷 < 0, then the function goes up in some directions and goes down in others, like 𝑧 = 𝑥2 −𝑦2 .
We say the function has a saddle point, that is, a critical point at which the function value
increases in some directions but decreases in others. (See Figure 15.15.)
• If 𝐷 = 0, then the quadratic function is 𝑎(𝑥 + 𝑏𝑦∕2𝑎)2 , whose graph is a parabolic cylinder. (See
Figure 15.16.)
 15.1 CRITICAL POINTS: LOCAL EXTREMA AND SADDLE POINTS 811

Figure 15.13: Local minimum: Figure 15.14: Local maximum: Figure 15.15: Saddle point: Figure 15.16: Parabolic
𝐷 > 0 and 𝑎 > 0 𝐷 > 0 and 𝑎 < 0 𝐷<0 cylinder: 𝐷 = 0
More generally, the graph of 𝑔(𝑥, 𝑦) = 𝑎(𝑥 − 𝑥0 )2 + 𝑏(𝑥 − 𝑥0 )(𝑦 − 𝑦0 ) + 𝑐(𝑦 − 𝑦0 )2 has the same
shape as the graph of 𝑓 (𝑥, 𝑦) = 𝑎𝑥2 + 𝑏𝑥𝑦 + 𝑐𝑦2 , except that the critical point is at (𝑥0 , 𝑦0 ) rather
than (0, 0).1
Classifying the Critical Points of a Function
Suppose that 𝑓 is any function with grad 𝑓 (0, 0) = 0⃗ . Its quadratic Taylor polynomial near (0, 0),
𝑓 (𝑥, 𝑦) ≈ 𝑓 (0, 0) + 𝑓𝑥 (0, 0)𝑥 + 𝑓𝑦 (0, 0)𝑦
1 1
+ 𝑓𝑥𝑥 (0, 0)𝑥2 + 𝑓𝑥𝑦 (0, 0)𝑥𝑦 + 𝑓𝑦𝑦 (0, 0)𝑦2,
2 2
can be simplified using 𝑓𝑥 (0, 0) = 𝑓𝑦 (0, 0) = 0, which gives
1 1
𝑓 (0, 0)𝑥2 + 𝑓𝑥𝑦 (0, 0)𝑥𝑦 + 𝑓𝑦𝑦 (0, 0)𝑦2.
𝑓 (𝑥, 𝑦) − 𝑓 (0, 0) ≈
2 𝑥𝑥 2
The discriminant of this quadratic polynomial is
1 1
( )( ) ( )2
𝐷 = 4𝑎𝑐 − 𝑏2 = 4 𝑓𝑥𝑥 (0, 0) 𝑓𝑦𝑦 (0, 0) − 𝑓𝑥𝑦 (0, 0) ,
2 2
which simplifies to
𝐷 = 𝑓𝑥𝑥 (0, 0)𝑓𝑦𝑦(0, 0) − (𝑓𝑥𝑦 (0, 0))2 .
There is a similar formula for 𝐷 if the critical point is at (𝑥0 , 𝑦0 ). An analogy with quadratic functions
suggests the following test for classifying a critical point of a function of two variables:

Second-Derivative Test for Functions of Two Variables

Suppose (𝑥0 , 𝑦0 ) is a point where grad 𝑓 (𝑥0 , 𝑦0 ) = 0⃗ . Let

𝐷 = 𝑓𝑥𝑥 (𝑥0 , 𝑦0 )𝑓𝑦𝑦 (𝑥0 , 𝑦0 ) − (𝑓𝑥𝑦 (𝑥0 , 𝑦0 ))2 .

• If 𝐷 > 0 and 𝑓𝑥𝑥 (𝑥0 , 𝑦0 ) > 0, then 𝑓 has a local minimum at (𝑥0 , 𝑦0 ).
• If 𝐷 > 0 and 𝑓𝑥𝑥 (𝑥0 , 𝑦0 ) < 0, then 𝑓 has a local maximum at (𝑥0 , 𝑦0 ).
• If 𝐷 < 0, then 𝑓 has a saddle point at (𝑥0 , 𝑦0 ).
• If 𝐷 = 0, anything can happen: 𝑓 can have a local maximum, or a local minimum, or a
saddle point, or none of these, at (𝑥0 , 𝑦0 ).

Example 6 Find the local maxima, minima, and saddle points of 𝑓 (𝑥, 𝑦) = 12 𝑥2 + 3𝑦3 + 9𝑦2 − 3𝑥𝑦 + 9𝑦 − 9𝑥.

Solution Setting the partial derivatives of 𝑓 to zero gives

𝑓𝑥 (𝑥, 𝑦) = 𝑥 − 3𝑦 − 9 = 0,
𝑓𝑦 (𝑥, 𝑦) = 9𝑦2 + 18𝑦 − 3𝑥 + 9 = 0.

1 We assumed that 𝑎 ≠ 0. If 𝑎 = 0 and 𝑐 ≠ 0, the same argument works. If both 𝑎 = 0 and 𝑐 = 0, then 𝑓 (𝑥, 𝑦) = 𝑏𝑥𝑦,

which has a saddle point.

812 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

Eliminating 𝑥 gives 9𝑦2 + 9𝑦 − 18 = 0, with solutions 𝑦 = −2 and 𝑦 = 1. The corresponding values

of 𝑥 are 𝑥 = 3 and 𝑥 = 12, so the critical points of 𝑓 are (3, −2) and (12, 1). The discriminant is
2
𝐷(𝑥, 𝑦) = 𝑓𝑥𝑥 𝑓𝑦𝑦 − 𝑓𝑥𝑦 = (1)(18𝑦 + 18) − (−3)2 = 18𝑦 + 9.

Since 𝐷(3, −2) = −36 + 9 < 0, we know that (3, −2) is a saddle point of 𝑓 . Since 𝐷(12, 1) =
18 + 9 > 0 and 𝑓𝑥𝑥 (12, 1) = 1 > 0, we know that (12, 1) is a local minimum of 𝑓 .

The second-derivative test does not give any information if 𝐷 = 0. However, as the following
example illustrates, we may still be able to classify the critical points.

Example 7 Classify the critical points of 𝑓 (𝑥, 𝑦) = 𝑥4 + 𝑦4 , and 𝑔(𝑥, 𝑦) = −𝑥4 − 𝑦4 , and ℎ(𝑥, 𝑦) = 𝑥4 − 𝑦4 .

Solution Each of these functions has a critical point at (0, 0). Since all the second partial derivatives are 0
there, each function has 𝐷 = 0. Near the origin, the graphs of 𝑓 , 𝑔 and ℎ look like the surfaces in
Figures 15.13–15.15, respectively, so 𝑓 has a local minimum at (0, 0), and 𝑔 has a local maximum
at (0, 0), and ℎ is saddle-shaped at (0, 0).
We can get the same results algebraically. Since 𝑓 (0, 0) = 0 and 𝑓 (𝑥, 𝑦) > 0 elsewhere, 𝑓 has
a local minimum at the origin. Since 𝑔(0, 0) = 0 and 𝑔(𝑥, 𝑦) < 0 elsewhere, 𝑔 has a local maximum
at the origin. Lastly, ℎ is saddle-shaped at the origin since ℎ(0, 0) = 0 and, away from the origin,
ℎ(𝑥, 𝑦) > 0 on the 𝑥-axis and ℎ(𝑥, 𝑦) < 0 on the 𝑦-axis.

Exercises and Problems for Section 15.1

EXERCISES

1. Figures (I)–(VI) show level curves of six functions 2. Which of the points 𝐴, 𝐵, 𝐶 in Figure 15.17 appear to
around a critical point 𝑃 . Does each function have a lo- be critical points? Classify those that are critical points.
cal maximum, a local minimum, or a saddle point at 𝑃 ?
(I) (II) 𝑦

2
1
𝑃 𝑃 0

𝐸 𝐺
1 𝐹
−5
−7 −2 −1 −1 −2
2
0
(III) (IV)
𝑥
−2 2 1
−1 1
2
1 1
1 0𝑃 1 −1 0𝑃 −1 𝐴 𝐷
2 𝐵 2
𝐶
−1 1
−2 2

−2 −1 0 −2
(V) (VI)

Figure 15.17
𝑃 𝑃
3. Which of the points 𝐷–𝐺 in Figure 15.17 appear to be
−1 (a) Local maxima?
12 (b) Local minima?
−2 14 (c) Saddle points?
 15.1 CRITICAL POINTS: LOCAL EXTREMA AND SADDLE POINTS 813

4. A function 𝑓 (𝑥, 𝑦) has partial derivatives 𝑓𝑥 (1, 2) = 3, In Exercises 14–27, find the critical points and classify them
𝑓𝑦 (1, 2) = 5. Explain how you know that 𝑓 does not as local maxima, local minima, saddle points, or none of
have a minimum at (1, 2). these.
5. Assume 𝑓 (𝑥, 𝑦) has a critical point at (2, 3) with 14. 𝑓 (𝑥, 𝑦) = 𝑥2 − 2𝑥𝑦 + 3𝑦2 − 8𝑦
𝑓 (2, 3) = 5. Draw possible cross-sections for 𝑥 = 2 15. 𝑓 (𝑥, 𝑦) = 5 + 6𝑥 − 𝑥2 + 𝑥𝑦 − 𝑦2
and 𝑦 = 3, and label one value on each axis, if 𝑓 (2, 3)
16. 𝑓 (𝑥, 𝑦) = 𝑥2 − 𝑦2 + 4𝑥 + 2𝑦
is:
17. 𝑓 (𝑥, 𝑦) = 400 − 3𝑥2 − 4𝑥 + 2𝑥𝑦 − 5𝑦2 + 48𝑦
(a) A local minimum
(b) A local maximum 18. 𝑓 (𝑥, 𝑦) = 15 − 𝑥2 + 2𝑦2 + 6𝑥 − 8𝑦
(c) A saddle point 19. 𝑓 (𝑥, 𝑦) = 𝑥2 𝑦 + 2𝑦2 − 2𝑥𝑦 + 6
20. 𝑓 (𝑥, 𝑦) = 2𝑥3 − 3𝑥2 𝑦 + 6𝑥2 − 6𝑦2
In Exercises 6–13, the function has a critical point at (0, 0).
What sort of critical point is it? 21. 𝑓 (𝑥, 𝑦) = 𝑥3 − 3𝑥 + 𝑦3 − 3𝑦
22. 𝑓 (𝑥, 𝑦) = 𝑥3 + 𝑦3 − 3𝑥2 − 3𝑦 + 10
6. 𝑓 (𝑥, 𝑦) = 𝑥2 − cos 𝑦 7. 𝑓 (𝑥, 𝑦) = 𝑥 sin 𝑦
23. 𝑓 (𝑥, 𝑦) = 𝑥3 + 𝑦3 − 6𝑦2 − 3𝑥 + 9
8. 𝑔(𝑥, 𝑦) = 𝑥4 + 𝑦3 9. 𝑓 (𝑥, 𝑦) = 𝑥6 + 𝑦6 24. 𝑓 (𝑥, 𝑦) = (𝑥 + 𝑦)(𝑥𝑦 + 1)
10. 𝑘(𝑥, 𝑦) = sin 𝑥 sin 𝑦 11. ℎ(𝑥, 𝑦) = cos 𝑥 cos 𝑦 25. 𝑓 (𝑥, 𝑦) = 8𝑥𝑦 − 14 (𝑥 + 𝑦)4
√
12. 𝑔(𝑥, 𝑦) = (𝑥 − 𝑒𝑥 )(1 − 𝑦2 ) 26. 𝑓 (𝑥, 𝑦) = 3 𝑥2 + 𝑦2
13. ℎ(𝑥, 𝑦) = 𝑥2 − 𝑥𝑦 + sin2 𝑦 27. 𝑓 (𝑥, 𝑦) = 𝑒2𝑥
2 +𝑦2

PROBLEMS
28. Let 𝑓 (𝑥, 𝑦) = 3𝑥2 + 𝑘𝑦2 + 9𝑥𝑦. Determine the values of
𝑘 (if any) for which the critical point at (0, 0) is:
−3
(a) A saddle point −2
−1
(b) A local maximum 3 2 1 0 0

(c) A local minimum

3 𝑇
3 2
𝑆 2
29. Let 𝑓 (𝑥, 𝑦) = 𝑥 + 𝑘𝑦 − 5𝑥𝑦. Determine the values of
1
𝑘 (if any) for which the critical point at (0, 0) is:
−3 −2 −1 −1 −2 −3
(a) A saddle point 𝑅 𝑄
(b) A local maximum
2
(c) A local minimum
3

30. Find 𝐴 and 𝐵 so that 𝑓 (𝑥, 𝑦) = 𝑥2 + 𝐴𝑥 + 𝑦2 + 𝐵 has a 0 0 1 2 3

local minimum value of 20 at (1, 0). 𝑃 −1
−2
−3
31. For 𝑓 (𝑥, 𝑦) = 𝑥2 + 𝑥𝑦 + 𝑦2 + 𝑎𝑥 + 𝑏𝑦 + 𝑐, find values of
𝑎, 𝑏, and 𝑐 giving a local minimum at (2, 5) and so that
𝑓 (2, 5) = 11.
2
32. (a) Find critical points for 𝑓 (𝑥, 𝑦) = 𝑒−(𝑥−𝑎) −(𝑦−𝑏) .
2
Figure 15.18
(b) Find 𝑎 and 𝑏 such that the critical point is at (−1, 5).
(c) For the values of 𝑎 and 𝑏 in part (b), is (−1, 5) a lo-
cal maximum, local minimum, or a saddle point? 34. Decide whether you think each point is a local maxi-
mum, local minimum, saddle point, or none of these.
33. Let 𝑓 (𝑥, 𝑦) = 𝑘𝑥2 + 𝑦2 − 4𝑥𝑦. Determine the values of
𝑘 (if any) for which the critical point at (0, 0) is:
(a) 𝑃 (b) 𝑄 (c) 𝑅 (d) 𝑆
(a) A saddle point
(b) A local maximum 35. Sketch the direction of ∇𝑓 at points surrounding each
(c) A local minimum of 𝑃 , 𝑅, 𝑆, and 𝑇 .
36. At which of 𝑃 , 𝑄, 𝑅, 𝑆, or 𝑇 does ‖∇𝑓 ‖ seem
For Problems 34–36, use the contours of 𝑓 in Figure 15.18. largest?.
814 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

For Problems 37–40, find critical points and classify them as 50. On a computer, draw contour diagrams for functions
local maxima, local minima, saddle points, or none of these.
𝑓 (𝑥, 𝑦) = 𝑘(𝑥2 + 𝑦2 ) − 2𝑥𝑦
2
37. 𝑓 (𝑥, 𝑦) = 𝑥3 + 𝑒−𝑦 for 𝑘 = −2, −1, 0, 1, 2. Use these figures to classify the
38. 𝑓 (𝑥, 𝑦) = sin 𝑥 sin 𝑦 critical point at (0, 0) for each value of 𝑘. Explain your
observations using the discriminant, 𝐷.
39. 𝑓 (𝑥, 𝑦) = 1 − cos 𝑥 + 𝑦2 ∕2
51. The behavior of a function can be complicated near a
40. 𝑓 (𝑥, 𝑦) = 𝑒𝑥 (1 − cos 𝑦) critical point where 𝐷 = 0. Suppose that
41. At the point (1, 3), suppose that 𝑓𝑥 = 𝑓𝑦 = 0 and
𝑓𝑥𝑥 > 0, 𝑓𝑦𝑦 > 0, 𝑓𝑥𝑦 = 0. 𝑓 (𝑥, 𝑦) = 𝑥3 − 3𝑥𝑦2 .

(a) What can you conclude about the behavior of the Show that there is one critical point at (0, 0) and that
function near the point (1, 3)? 𝐷 = 0 there. Show that the contour for 𝑓 (𝑥, 𝑦) = 0
(b) Sketch a possible contour diagram. consists of three lines intersecting at the origin and that
these lines divide the plane into six regions around the
42. At the point (𝑎, 𝑏), suppose that 𝑓𝑥 = 𝑓𝑦 = 0, 𝑓𝑥𝑥 > 0, origin where 𝑓 alternates from positive to negative.
𝑓𝑦𝑦 = 0, 𝑓𝑥𝑦 > 0. Sketch a contour diagram for 𝑓 near (0, 0). The graph
(a) What can you conclude about the shape of the of this function is called a monkey saddle.
graph of 𝑓 near the point (𝑎, 𝑏)? 52. The contour diagrams for four functions 𝑧 = 𝑓 (𝑥, 𝑦)
(b) Sketch a possible contour diagram. are in (a)–(d). Each function has a critical point with
𝑧 = 0 at the origin. Graphs (I)–(IV) show the value of
43. Let ℎ(𝑥, 𝑦) = 𝑓 (𝑥)𝑔(𝑦) where 𝑓 (0) = 𝑔(0) = 0 and
𝑧 for these four functions on a small circle around the
𝑓 ′ (0) ≠ 0, 𝑔 ′ (0) ≠ 0. Show that (0, 0) is a saddle point
origin, expressed as function of 𝜃, the angle between
of ℎ.
the positive 𝑥-axis and a line through the origin. Match
44. Let ℎ(𝑥, 𝑦) = 𝑓 (𝑥) + 𝑔(𝑦). Show that ℎ has a critical the contour diagrams (a)–(d) with the graphs (I)–(IV).
point at (𝑎, 𝑏) if 𝑓 ′ (𝑎) = 𝑔 ′ (𝑏) = 0, and, assuming Classify the critical points as local maxima, local min-
𝑓 ′′ (𝑎) ≠ 0 and 𝑔 ′′ (𝑏) ≠ 0, it is a local maximum or min- ima or saddle points.
imum when 𝑓 ′′ (𝑎) and 𝑔 ′′ (𝑏) have the same sign and a
saddle point when they have opposite signs.
(a) 𝑦 (b) 𝑦
45. Let ℎ(𝑥, 𝑦) = (𝑓 (𝑥))2 + (𝑔(𝑦))2 . Show that if 𝑓 (𝑎) =
𝑔(𝑏) = 0, then (𝑎, 𝑏) is a local minimum.
46. Draw a possible contour diagram of 𝑓 such that
𝑥 𝑥
𝑓𝑥 (−1, 0) = 0, 𝑓𝑦 (−1, 0) < 0, 𝑓𝑥 (3, 3) > 0, 𝑓𝑦 (3, 3) >
0, and 𝑓 has a local maximum at (3, −3).
47. Draw a possible contour diagram of a function with a
saddle point at (2, 1), a local minimum at (2, 4), and no (c) 𝑦 (d) 𝑦
other critical points. Label the contours.
48. For constants 𝑎 and 𝑏 with 𝑎𝑏 ≠ 0 and 𝑎𝑏 ≠ 1, let
𝑥 𝑥
𝑓 (𝑥, 𝑦) = 𝑎𝑥2 + 𝑏𝑦2 − 2𝑥𝑦 − 4𝑥 − 6𝑦.

(a) Find the 𝑥- and 𝑦-coordinates of the critical point.

Your answer will be in terms of 𝑎 and 𝑏.
(b) If 𝑎 = 𝑏 = 2, is the critical point a local maxi-
mum, a local minimum, or neither? Give a reason (I) 𝑧 (II) 𝑧
for your answer.
(c) Classify the critical point for all values of 𝑎 and 𝑏 𝜃 𝜃
with 𝑎𝑏 ≠ 0 and 𝑎𝑏 ≠ 1. 𝜋 2𝜋 𝜋 2𝜋

49. (a) Find the critical point of 𝑓 (𝑥, 𝑦) = (𝑥2 − 𝑦)(𝑥2 + 𝑦). (III) 𝑧 (IV) 𝑧
(b) Show that at the critical point, the discriminant
𝐷 = 0, so the second-derivative test gives no in-
𝜃 𝜃
formation about the nature of the critical point. 𝜋 2𝜋 𝜋 2𝜋
(c) Sketch contours near the critical point to determine
whether it is a local maximum, a local minimum,
a saddle point, or none of these.
 15.2 OPTIMIZATION 815

Strengthen Your Understanding

In Problems 53–55, explain what is wrong with the state- 0.7

ment. 0.5
0.2
53. If 𝑓𝑥 = 𝑓𝑦 = 0 at (1, 3), then 𝑓 has a local maximum or
0.05
local minimum at (1, 3).
54. For 𝑓 (𝑥, 𝑦), if 𝐷 = 𝑓𝑥𝑥 𝑓𝑦𝑦 − (𝑓𝑥𝑦 )2 = 0 at (𝑎, 𝑏), then 0 0
0.5 0.5
(𝑎, 𝑏) is a saddle point. 0.2 0.05 0.05 0.2
0.7 𝑃 0.7
55. A critical point (𝑎, 𝑏) for the function 𝑓 must be a local 0 0
minimum if both cross-sections for 𝑥 = 𝑎 and 𝑦 = 𝑏 are
concave up. 0.05
0.2
In Problems 56–57, give an example of: 0.5
0.7
56. A nonlinear function having no critical points
57. A function 𝑓 (𝑥, 𝑦) with a local maximum at (2, −3, 4). Figure 15.19
√
Are the statements in Problems 58–69 true or false? Give 63. The function 𝑓 (𝑥, 𝑦) = 𝑥2 + 𝑦2 has a local minimum
reasons for your answer. at the origin.
64. The function 𝑓 (𝑥, 𝑦) = 𝑥2 − 𝑦2 has a local minimum at
58. If 𝑓𝑥 (𝑃0 ) = 𝑓𝑦 (𝑃0 ) = 0, then 𝑃0 is a critical point of 𝑓 . the origin.
59. If 𝑓𝑥 (𝑃0 ) = 𝑓𝑦 (𝑃0 ) = 0, then 𝑃0 is a local maximum or 65. If 𝑓 has a local minimum at 𝑃0 then so does the function
local minimum of 𝑓 . 𝑔(𝑥, 𝑦) = 𝑓 (𝑥, 𝑦) + 5.
60. If 𝑃0 is a critical point of 𝑓 , then 𝑃0 is either a local 66. If 𝑓 has a local minimum at 𝑃0 then the function
maximum or local minimum of 𝑓 . 𝑔(𝑥, 𝑦) = −𝑓 (𝑥, 𝑦) has a local maximum at 𝑃0 .
61. If 𝑃0 is a local maximum or local minimum of 𝑓 , and 67. Every function has at least one local maximum.
not on the boundary of the domain of 𝑓 , then 𝑃0 is a 68. If 𝑃0 is a local maximum of 𝑓 , then 𝑓 (𝑎, 𝑏) ≤ 𝑓 (𝑃0 ) for
critical point of 𝑓 . all points (𝑎, 𝑏) in 2-space.
62. The function whose contour diagram is shown in Fig- 69. If 𝑃0 is a local maximum of 𝑓 , then 𝑃0 is also a global
ure 15.19 has a saddle point at 𝑃 . maximum of 𝑓 .

15.2 OPTIMIZATION

Suppose we want to find the highest and the lowest points in Colorado. A contour map is shown
in Figure 15.20. The highest point is the top of a mountain peak (point A on the map, Mt. Elbert,
14,440 feet high). What about the lowest point? Colorado does not have large pits without drainage,
like Death Valley in California. A drop of rain falling at any point in Colorado will eventually flow
out of the state. If there is no local minimum inside the state, where is the lowest point? It must be
on the state boundary at a point where a river is flowing out of the state (point B where the Arikaree
River leaves the state, 3,315 feet high). The highest point in Colorado is a global maximum for the
elevation function in Colorado and the lowest point is the global minimum.
R.
N.

tte
Pla

Pla
tte

uth
R.

Longs So
Peak
Denver .
.
Grays Peak e eR B
oR kar
Ari
R.

rad
olo
latte

C
Mt. Elbert
th P

A
Sou

Grand Junction

Arka
nsas
R.
Mt. Wilson Pueblo
Lamar
R.
re
toi
rga

Alamosa
Pu

Figure 15.20: The highest and lowest points in the state of Colorado
816 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

In general, if we are given a function 𝑓 defined on a region 𝑅, we say:

• 𝑓 has a global maximum on 𝑅 at the point 𝑃0 if 𝑓 (𝑃0 ) ≥ 𝑓 (𝑃 ) for all points 𝑃 in 𝑅.

• 𝑓 has a global minimum on 𝑅 at the point 𝑃0 if 𝑓 (𝑃0 ) ≤ 𝑓 (𝑃 ) for all points 𝑃 in 𝑅.

The process of finding a global maximum or minimum for a function 𝑓 on a region 𝑅 is called
optimization. If the region 𝑅 is not stated explicitly, we take it to be the whole 𝑥𝑦-plane unless the
context of the problem suggests otherwise.
How Do We Find Global Maxima and Minima?
As the Colorado example illustrates, a global extremum can occur either at a critical point inside
the region or at a point on the boundary of the region. This is analogous to single-variable calculus,
where a function achieves its global extrema on an interval either at a critical point inside the interval
or at an endpoint of the interval.

To locate global maxima and minima for a function 𝑓 on a region 𝑅:

• Find the critical points of 𝑓 in the region 𝑅.
• Investigate whether the critical points give global maxima or minima.
• If the region 𝑅 has a boundary, investigate whether 𝑓 attains a global maximum or min-
imum on the boundary of 𝑅.

Investigating the boundary of a region for possible maxima and minima is the topic of Sec-
tion 15.1. In this section, we focus on finding global maxima and minima of functions on regions
that do not include boundaries.
Not all functions have a global maximum or minimum: it depends on the function and the region.
First, we consider applications in which global extrema are expected from practical considerations.
At the end of this section, we examine the conditions that lead to global extrema. In general, the
fact that a function has a single local maximum or minimum does not guarantee that the point is
the global maximum or minimum. (See Problem 38.) An exception is if the function is quadratic, in
which case the local maximum or minimum is the global maximum or minimum. (See Example 1
on page 807 and Example 5 on page 810.)
Maximizing Profit and Minimizing Cost
In planning production of an item, a company often chooses the combination of price and quantity
that maximizes its profit. We use
Profit = Revenue − Cost,
and, provided the price is constant,
Revenue = Price ⋅ Quantity = 𝑝𝑞.
In addition, we need to know how the cost and price depend on quantity.

Example 1 A company manufactures two items which are sold in two separate markets where it has a monopoly.
The quantities, 𝑞1 and 𝑞2 , demanded by consumers, and the prices, 𝑝1 and 𝑝2 (in dollars), of each
item are related by
𝑝1 = 600 − 0.3𝑞1 and 𝑝2 = 500 − 0.2𝑞2 .
Thus, if the price for either item increases, the demand for it decreases. The company’s total produc-
tion cost is given by
𝐶 = 16 + 1.2𝑞1 + 1.5𝑞2 + 0.2𝑞1 𝑞2 .
To maximize its total profit, how much of each product should be produced? What is the maximum
profit? 2
2 Adapted from M. Rosser and P. Lis, Basic Mathematics for Economists, 3rd ed. (New York: Routledge, 2016), p. 351.
 15.2 OPTIMIZATION 817

Solution The total revenue, 𝑅, is the sum of the revenues, 𝑝1 𝑞1 and 𝑝2 𝑞2 , from each market. Substituting for
𝑝1 and 𝑝2 , we get

𝑅 = 𝑝1 𝑞1 + 𝑝2 𝑞2 = (600 − 0.3𝑞1 )𝑞1 + (500 − 0.2𝑞2 )𝑞2

= 600𝑞1 − 0.3𝑞12 + 500𝑞2 − 0.2𝑞22 .

Thus, the total profit 𝑃 is given by

𝑃 = 𝑅 − 𝐶 = 600𝑞1 − 0.3𝑞12 + 500𝑞2 − 0.2𝑞22 − (16 + 1.2𝑞1 + 1.5𝑞2 + 0.2𝑞1 𝑞2 )

= −16 + 598.8𝑞1 − 0.3𝑞12 + 498.5𝑞2 − 0.2𝑞22 − 0.2𝑞1 𝑞2 .

Since 𝑞1 and 𝑞2 cannot be negative,3 the region we consider is the first quadrant with boundary 𝑞1 = 0
and 𝑞2 = 0.
To maximize 𝑃 , we look for critical points by setting the partial derivatives equal to 0:
𝜕𝑃
= 598.8 − 0.6𝑞1 − 0.2𝑞2 = 0,
𝜕𝑞1
𝜕𝑃
= 498.5 − 0.4𝑞2 − 0.2𝑞1 = 0.
𝜕𝑞2

Since grad 𝑃 is defined everywhere, the only critical points of 𝑃 are those where grad 𝑃 = 0⃗ . Thus,
solving for 𝑞1 , and 𝑞2 , we find that

𝑞1 = 699.1 and 𝑞2 = 896.7.

The corresponding prices are

𝑝1 = 390.27 and 𝑝2 = 320.66.

To see whether or not we have found a local maximum, we compute second partial derivatives:

𝜕2 𝑃 𝜕2 𝑃 𝜕2 𝑃
= −0.6, = −0.4, = −0.2,
𝜕𝑞12 𝜕𝑞22 𝜕𝑞1 𝜕𝑞2

so,
( )2
𝜕2 𝑃 𝜕2 𝑃 𝜕2 𝑃
𝐷= − = (−0.6)(−0.4) − (−0.2)2 = 0.2.
𝜕𝑞12 𝜕𝑞22 𝜕𝑞1 𝜕𝑞2
Therefore we have found a local maximum. The graph of 𝑃 is a paraboloid opening downward, so
(699.1, 896.7) is a global maximum. This point is within the region, so points on the boundary give
smaller values of 𝑃 .
The company should produce 699.1 units of the first item priced at $390.27 per unit, and 896.7
units of the second item priced at $320.66 per unit. The maximum profit 𝑃 (699.1, 896.7) ≈ $433,000.

Example 2 A delivery of 480 cubic meters of gravel is to be made to a landfill. The trucker plans to purchase an
open-top box in which to transport the gravel in numerous trips. The total cost to the trucker is the
cost of the box plus $80 per trip. The box must have height 2 meters, but the trucker can choose the
length and width. The cost of the box is $100/m2 for the ends, $50/m2 for the sides and $200/m2 for
the bottom. Notice the tradeoff: A smaller box is cheaper to buy but requires more trips. What size
box should the trucker buy to minimize the total cost?4
3 Restricting prices to be nonnegative further restricts the region but does not alter the solution.
4 Adapted from Claude McMillan, Jr., Mathematical Programming, 2nd ed. (New York: Wiley, 1978), pp. 156–157.
818 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

Solution We first get an algebraic expression for the trucker’s cost. Let the length of the box be 𝑥 meters and
the width be 𝑦 meters; the height is 2 meters. (See Figure 15.21.)

Table 15.2 Trucker’s itemized cost

Expense Cost in dollars

✛
2m Travel: 480∕(2𝑥𝑦) at $80/trip 80 ⋅ 480∕(2𝑥𝑦)

✛
2 2
Ends: 2 at $100/m ⋅ 2𝑦 m 400𝑦
Sides: 2 at $50/m2 ⋅ 2𝑥 m2 200𝑥
𝑦 𝑥 Bottom: l at $200/m2 ⋅ 𝑥𝑦 m2 200𝑥𝑦

Figure 15.21: The box for transporting gravel

The volume of the box is 2𝑥𝑦 m3 , so delivery of 480 m3 of gravel requires 480∕(2𝑥𝑦) trips. The
number of trips is a whole number; however, we treat it as continuous so that we can optimize using
derivatives. The trucker’s cost is itemized in Table 15.2. The problem is to minimize
( )
480 96
Total cost = 80 ⋅ + 400𝑦 + 200𝑥 + 200𝑥𝑦 = 200 + 2𝑦 + 𝑥 + 𝑥𝑦 .
2𝑥𝑦 𝑥𝑦
The length and width of the box must be positive. Thus, the region is the first quadrant but it does
not contain the boundary, 𝑥 = 0 and 𝑦 = 0.
Our problem is to minimize
96
𝑓 (𝑥, 𝑦) = + 2𝑦 + 𝑥 + 𝑥𝑦.
𝑥𝑦
The critical points of this function occur where
96
𝑓𝑥 (𝑥, 𝑦) = − +1+𝑦=0
𝑥2 𝑦
96
𝑓𝑦 (𝑥, 𝑦) = − + 2 + 𝑥 = 0.
𝑥𝑦2
We put the 96∕(𝑥2 𝑦) and 96∕(𝑥𝑦2 ) terms on the other side of the the equation, divide, and simplify:
96∕(𝑥2 𝑦) 1+𝑦 𝑦 1+𝑦
= so = giving 2𝑦 = 𝑥.
96∕(𝑥𝑦2 ) 2 + 𝑥 𝑥 2+𝑥
Substituting 𝑥 = 2𝑦 in the equation 𝑓𝑦 (𝑥, 𝑦) = 0 gives
96
− + 2 + 2𝑦 = 0
2𝑦 ⋅ 𝑦2
𝑦4 + 𝑦3 − 24 = 0.
The only positive solution to this equation is 𝑦 = 2, so the only critical point in the region is (4, 2).
To check that the critical point is a local minimum, we use the second-derivative test. Since
)2
192 192 96 25
(
𝐷(4, 2) = 𝑓𝑥𝑥 𝑓𝑦𝑦 − (𝑓𝑥𝑦 )2 = 3 ⋅ − + 1 =9− >0
4 ⋅ 2 4 ⋅ 23 42 ⋅ 22 4
and 𝑓𝑥𝑥 (4, 2) > 0, the point (4, 2) is a local minimum. Since the value of 𝑓 increases without bound
as 𝑥 or 𝑦 increases without bound and as 𝑥 → 0+ and 𝑦 → 0+ , it can be shown that (4, 2) is a global
minimum. (See Problem 42.) Thus, the optimal box is 4 meters long and 2 meters wide. With a box
of this size, the trucker would need to make 30 trips to haul all of the gravel. This large number lends
some credibility to our decision to treat the number of trips as a continuous variable.
 15.2 OPTIMIZATION 819

Fitting a Line to Data: Least Squares

Suppose we want to fit the “best” line to some data in the plane. We measure the distance from a
line to the data points by adding the squares of the vertical distances from each point to the line. The
smaller this sum of squares is, the better the line fits the data. The line with the minimum sum of
square distances is called the least squares line, or the regression line. If the data is nearly linear,
the least squares line is a good fit; otherwise it may not be. (See Figure 15.22.)

Data almost linear: line fits well Data not very linear: line does not fit well

Figure 15.22: Fitting lines to data points

Example 3 Find a least squares line for the following data points: (1, 1), (2, 1), and (3, 3).

Solution Suppose the line has equation 𝑦 = 𝑏 + 𝑚𝑥. If we find 𝑏 and 𝑚 then we have found the line. So,
for this problem, 𝑏 and 𝑚 are the two variables. Any values of 𝑚 and 𝑏 are possible, so this is an
unconstrained problem. We want to minimize the function 𝑓 (𝑏, 𝑚) that gives the sum of the three
squared vertical distances from the points to the line in Figure 15.23.
𝑦

3 (3, 3) 𝑦 = 𝑏 + 𝑚𝑥
❄
✻
(3, 𝑏 + 3𝑚)

2
(2, 𝑏 + 2𝑚)
✻
(1, 1) ❄
(2, 1)
1
❄
✻
(1, 𝑏 + 𝑚)

0 𝑥
1 2 3

Figure 15.23: The least squares line minimizes the sum of the squares of these vertical distances

The vertical distance from the point (1, 1) to the line is the difference in the 𝑦-coordinates 1 −
(𝑏 + 𝑚); similarly for the other points. Thus, the sum of squares is
𝑓 (𝑏, 𝑚) = (1 − (𝑏 + 𝑚))2 + (1 − (𝑏 + 2𝑚))2 + (3 − (𝑏 + 3𝑚))2 .
To minimize 𝑓 we look for critical points. First we differentiate 𝑓 with respect to 𝑏:
𝜕𝑓
= −2(1 − (𝑏 + 𝑚)) − 2(1 − (𝑏 + 2𝑚)) − 2(3 − (𝑏 + 3𝑚))
𝜕𝑏
= −2 + 2𝑏 + 2𝑚 − 2 + 2𝑏 + 4𝑚 − 6 + 2𝑏 + 6𝑚
= −10 + 6𝑏 + 12𝑚.
820 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

Now we differentiate with respect to 𝑚:

𝜕𝑓
= 2(1 − (𝑏 + 𝑚))(−1) + 2(1 − (𝑏 + 2𝑚))(−2) + 2(3 − (𝑏 + 3𝑚))(−3)
𝜕𝑚
= −2 + 2𝑏 + 2𝑚 − 4 + 4𝑏 + 8𝑚 − 18 + 6𝑏 + 18𝑚
= −24 + 12𝑏 + 28𝑚.
𝜕𝑓 𝜕𝑓
The equations = 0 and = 0 give a system of two linear equations in two unknowns:
𝜕𝑏 𝜕𝑚
−10 + 6𝑏 + 12𝑚 = 0,
−24 + 12𝑏 + 28𝑚 = 0.

The solution to this pair of equations is the critical point 𝑏 = −1∕3 and 𝑚 = 1. Since

𝐷 = 𝑓𝑏𝑏 𝑓𝑚𝑚 − (𝑓𝑚𝑏 )2 = (6)(28) − 122 = 24 and 𝑓𝑏𝑏 = 6 > 0,

we have found a local minimum. The graph of 𝑓 (𝑏, 𝑚) is a parabola opening upward, so the local
minimum is the global minimum of 𝑓 . Thus, the least squares line is

1
𝑦=𝑥− .
3
As a check, notice that the line 𝑦 = 𝑥 passes through the points (1, 1) and (3, 3). It is reasonable that
introducing the point (2, 1) moves the 𝑦-intercept down from 0 to −1∕3.

The general formulas for the slope and 𝑦-intercept of a least squares line are in Project 2 (avail-
able online). Many calculators have built-in formulas for 𝑏 and 𝑚, as well as for the correlation
coefficient, which measures how well the data points fit the least squares line.

How Do We Know Whether a Function Has a Global Maximum or Minimum?

Under what circumstances does a function of two variables have a global maximum or minimum?
The next example shows that a function may have both a global maximum and a global minimum
on a region, or just one, or neither.

Example 4 Investigate the global maxima and minima of the following functions:
(a) ℎ(𝑥, 𝑦) = 1 + 𝑥2 + 𝑦2 on the disk 𝑥2 + 𝑦2 ≤ 1.
(b) 𝑓 (𝑥, 𝑦) = 𝑥2 − 2𝑥 + 𝑦2 − 4𝑦 + 5 on the 𝑥𝑦-plane.
(c) 𝑔(𝑥, 𝑦) = 𝑥2 − 𝑦2 on the 𝑥𝑦-plane.

Solution (a) The graph of ℎ(𝑥, 𝑦) = 1 + 𝑥2 + 𝑦2 is a bowl-shaped paraboloid with a global minimum of 1 at
(0, 0), and a global maximum of 2 on the edge of the region, 𝑥2 + 𝑦2 = 1.
(b) The graph of 𝑓 in Figure 15.5 on page 807 shows that 𝑓 has a global minimum at the point (1, 2)
and no global maximum (because the value of 𝑓 increases without bound as 𝑥 → ∞, 𝑦 → ∞).
(c) The graph of 𝑔 in Figure 15.7 on page 808 shows that 𝑔 has no global maximum because
𝑔(𝑥, 𝑦) → ∞ as 𝑥 → ∞ if 𝑦 is constant. Similarly, 𝑔 has no global minimum because 𝑔(𝑥, 𝑦) →
−∞ as 𝑦 → ∞ if 𝑥 is constant.

Sometimes a function is guaranteed to have a global maximum and minimum. For example, a
continuous function, ℎ(𝑥), of one variable has a global maximum and minimum on every closed
interval 𝑎 ≤ 𝑥 ≤ 𝑏. On a non-closed interval, such as 𝑎 ≤ 𝑥 < 𝑏 or 𝑎 < 𝑥 < 𝑏, or on an unbounded
interval, such as 𝑎 < 𝑥 < ∞, ℎ may not have a maximum or minimum value.
 15.2 OPTIMIZATION 821

What is the situation for functions of two variables? As it turns out, a similar result is true for
continuous functions defined on regions which are closed and bounded, analogous to the closed and
bounded interval 𝑎 ≤ 𝑥 ≤ 𝑏. In everyday language we say

• A closed region is one which contains its boundary;

• A bounded region is one which does not stretch to infinity in any direction.

More precise definitions follow. Suppose 𝑅 is a region in 2-space. A point (𝑥0 , 𝑦0 ) is a boundary
point of 𝑅 if, for every 𝑟 > 0, the circular disk with center (𝑥0 , 𝑦0 ) and radius 𝑟 contains both
points which are in 𝑅 and points which are not in 𝑅. See Figure 15.24. A point (𝑥0 , 𝑦0 ) can be a
boundary point of the region 𝑅 without belonging to 𝑅. The collection of all the boundary points is
the boundary of 𝑅. The region 𝑅 is closed if it contains its boundary.
A region 𝑅 in 2-space is bounded if the distance between every point (𝑥, 𝑦) in 𝑅 and the origin
is less than some constant 𝐾. Closed and bounded regions in 3-space are defined in the same way.

√
Example 5 (a) Consider the square −1 ≤ 𝑥 ≤ 1, −1 ≤ 𝑦 ≤ 1. Every point in this region is within distance 2
of the origin, so the region is bounded. The region’s boundary consists of four line segments, all
of which belong to the region, so the region is closed.
(b) Consider the first quadrant 𝑥 ≥ 0, 𝑦 ≥ 0. The boundary of this region consists of the origin, the
positive 𝑥-axis, and the positive 𝑦-axis. All of these belong to the region, so the region is closed.
However, the region is not bounded, since there is no upper bound on distances between points
in the region and the origin.
(c) The disk 𝑥2 + 𝑦2 < 1 is bounded, because each point in the region is within distance 1 of the
origin. However, the disk is not closed, because (1, 0) is a boundary point of the region but not
included in the region.
(d) The half-plane 𝑦 > 0 is neither closed nor bounded. The origin is a boundary point of this region
but is not included in the region.

The reason that closed and bounded regions are useful is the following theorem, which is also
true for functions of three or more variables:5

Theorem 15.1: Extreme Value Theorem for Multivariable Functions

If 𝑓 is a continuous function on a closed and bounded region 𝑅, then 𝑓 has a global maximum
at some point (𝑥0 , 𝑦0 ) in 𝑅 and a global minimum at some point (𝑥1 , 𝑦1 ) in 𝑅.

If 𝑓 is not continuous or the region 𝑅 is not closed and bounded, there is no guarantee that 𝑓
achieves a global maximum or global minimum on 𝑅. In Example 4, the function 𝑔 is continuous
but does not achieve a global maximum or minimum in 2-space, a region which is closed but not
bounded. Example 6 illustrates what can go wrong when the region is bounded but not closed.

(𝑥0 , 𝑦0 )

Figure 15.24: Boundary point (𝑥0 , 𝑦0 ) of 𝑅

5 For a proof, see Walter Rudin, Principles of Mathematical Analysis, 3rd ed. (New York: McGraw-Hill, 1976), p. 89.
822 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

Example 6 Does the function 𝑓 have a global maximum or minimum on the region 𝑅 given by 0 < 𝑥2 + 𝑦2 ≤ 1?

1
𝑓 (𝑥, 𝑦) =
𝑥2 + 𝑦2
Solution The region 𝑅 is bounded, but it is not closed since it does not contain the boundary point (0, 0).
We see from the graph of 𝑧 = 𝑓 (𝑥, 𝑦) in Figure 15.25 that 𝑓 has a global minimum on the circle
𝑥2 + 𝑦2 = 1. However, 𝑓 (𝑥, 𝑦) → ∞ as (𝑥, 𝑦) → (0, 0), so 𝑓 has no global maximum.
𝑧

𝑥 𝑦
1
Figure 15.25: Graph showing 𝑓 (𝑥, 𝑦) = 𝑥2 +𝑦2
has no global maximum on 0 < 𝑥2 + 𝑦2 ≤ 1

Exercises and Problems for Section 15.2

EXERCISES
1. By looking at the weather map in Figure 12.1 on page 652, find the maximum and minimum daily high temperatures in
the states of Mississippi, Alabama, Pennsylvania, New York, California, Arizona, and Massachusetts.

In Exercises 2–4, estimate the position and approximate value of the global maxima and minima on the closed region shown.
2. 𝑦 3. 𝑦 4. 𝑦
6 5 5𝜋∕2
21
22
5
0.75
0.50
0.25
9 4 23 2𝜋
4 7 24
0 3
3 4 3 3𝜋∕2 0
25
2
−0.50

−0.75

−0.25
2 2 26 𝜋
1 6
27
4 𝑥 1 𝜋∕2 0
28
1 2 3 4 5 6 7 29 0.
𝑥 0.50 0.75 25

1 2 3 4 5 𝑥
𝜋∕2 𝜋

In Problems 5–9, without calculus, find the highest and low- In Exercises 11–13, find the global maximum and minimum
est points (if they exist) on the surface. The 𝑧-axis is upward. of the function on −1 ≤ 𝑥 ≤ 1, −1 ≤ 𝑦 ≤ 1, and say whether
it occurs on the boundary of the square. [Hint: Use graphs.]
5. 𝑥2 + 𝑦2 + (𝑧 − 1)2 = 49 11. 𝑧 = 𝑥2 + 𝑦2 12. 𝑧 = −𝑥2 −𝑦2 13. 𝑧 = 𝑥2 − 𝑦2
6. (𝑥 + 1)2 + (𝑦 − 3)2 + 2𝑧2 = 162
In Problems 14–21, does the function have a global maxi-
7. 𝑧 = (𝑥 − 5)2 + (𝑦 − 𝜋)2 + 2𝜋 mum? A global minimum?

8. 𝑧 = 44 − 2𝑥2 − 2𝑦2 14. 𝑓 (𝑥, 𝑦) = 𝑥2 − 2𝑦2 15. 𝑔(𝑥, 𝑦) = 𝑥2 𝑦2

3 3
9. 𝑥 = 4 + 𝑦2 + 2𝑧2 16. ℎ(𝑥, 𝑦) = 𝑥 + 𝑦 17. 𝑓 (𝑥, 𝑦) = −2𝑥2 − 7𝑦2
𝑥2 +𝑦2
18. 𝑓 (𝑥, 𝑦) = 𝑒 19. ℎ(𝑥, 𝑦) = 1 − 𝑦2 𝑒𝑥𝑦
10. The surface 𝑧 = 27 − 𝑥2 − 𝑦2 cuts the plane 𝑧 = 2 in
a curve. Without calculus, find the point on this curve 20. 𝑓 (𝑥, 𝑦) = 𝑥2 ∕2 + 3𝑦3 + 9𝑦2 − 3𝑥
with the greatest 𝑦-coordinate. 21. 𝑔(𝑥, 𝑦) = 𝑥2 − cos(𝑥 + 𝑦)
 15.2 OPTIMIZATION 823

PROBLEMS
22. (a) Compute and classify the critical points of (a) Find the maximum profit that can be made, assum-
𝑓 (𝑥, 𝑦) = 2𝑥2 − 3𝑥𝑦 + 8𝑦2 + 𝑥 − 𝑦. ing the prices are fixed.
(b) By completing the square, plot the contour diagram (b) Find the rate of change of that maximum profit as
of 𝑓 and show that the local extremum found in 𝑝1 increases.
part (a) is a global one.
35. A company operates two plants which manufacture the
In Problems 23–26, find and classify the critical points of same item and whose total cost functions are
the function. Then decide whether the function has global
extrema on the 𝑥𝑦-plane, and find them if they exist. 𝐶1 = 8.5 + 0.03𝑞12 and 𝐶2 = 5.2 + 0.04𝑞22 ,

23. 𝑓 (𝑥, 𝑦) = 𝑦2 + 2𝑥𝑦 − 𝑦 − 𝑥3 − 𝑥 + 2 where 𝑞1 and 𝑞2 are the quantities produced by each
24. 𝑓 (𝑥, 𝑦) = 2𝑥2 − 4𝑥𝑦 + 5𝑦2 + 9𝑦 + 2 plant. The company is a monopoly. The total quantity
25. 𝑓 (𝑥, 𝑦) = −𝑥4 + 2𝑥𝑦2 − 2𝑦3 demanded, 𝑞 = 𝑞1 + 𝑞2 , is related to the price, 𝑝, by
2 −𝑦2
26. 𝑓 (𝑥, 𝑦) = 𝑥𝑒−𝑥
𝑝 = 60 − 0.04𝑞.
27. A closed rectangular box has volume 32 cm3 . What are
the lengths of the edges giving the minimum surface How much should each plant produce in order to max-
area? imize the company’s profit?6
28. A closed rectangular box with faces parallel to the co-
36. The quantity of a product demanded by consumers is
ordinate planes has one bottom corner at the origin and
a function of its price. The quantity of one product de-
the opposite top corner in the first octant on the plane
manded may also depend on the price of other prod-
3𝑥 + 2𝑦 + 𝑧 = 1. What is the maximum volume of such
ucts. For example, if the only chocolate shop in town (a
a box?
monopoly) sells milk and dark chocolates, the price it
29. An international airline has a regulation that each pas- sets for each affects the demand of the other. The quan-
senger can carry a suitcase having the sum of its width, tities demanded, 𝑞1 and 𝑞2 , of two products depend on
length and height less than or equal to 135 cm. Find the their prices, 𝑝1 and 𝑝2 , as follows:
dimensions of the suitcase of maximum volume that a
passenger may carry under this regulation. 𝑞1 = 150 − 2𝑝1 − 𝑝2
30. Design a rectangular milk carton box of width 𝑤, length
𝑙, and height ℎ which holds 512 cm3 of milk. The sides 𝑞2 = 200 − 𝑝1 − 3𝑝2 .
of the box cost 1 cent∕cm2 and the top and bottom cost
2 cent∕cm2 . Find the dimensions of the box that mini- (a) What does the fact that the coefficients of 𝑝1 and
mize the total cost of materials used. 𝑝2 are negative tell you? Give an example of two
products that might be related this way.
31. Find the point on the plane 3𝑥+2𝑦+𝑧 = 1 that is closest (b) If one manufacturer sells both products, how
to the origin by minimizing the square of the distance. should the prices be set to generate the maximum
32. What is the shortest distance from the surface 𝑥𝑦+3𝑥+ possible revenue? What is that maximum possible
𝑧2 = 9 to the origin? revenue?
33. For constants 𝑎, 𝑏, and 𝑐, let 𝑓 (𝑥, 𝑦) = 𝑎𝑥 + 𝑏𝑦 + 𝑐 be a
linear function, and let 𝑅 be a region in the 𝑥𝑦-plane. 37. A company manufactures a product which requires cap-
ital and labor to produce. The quantity, 𝑄, of the product
(a) If 𝑅 is any disk, show that the maximum and min- manufactured is given by the Cobb-Douglas function
imum values of 𝑓 on 𝑅 occur on the boundary of
the disk. 𝑄 = 𝐴𝐾 𝑎 𝐿𝑏 ,
(b) If 𝑅 is any rectangle, show that the maximum and
minimum values of 𝑓 on 𝑅 occur at the corners of where 𝐾 is the quantity of capital; 𝐿 is the quantity of
the rectangle. They may occur at other points of the labor used; and 𝐴, 𝑎, and 𝑏 are positive constants with
rectangle as well. 0 < 𝑎 < 1 and 0 < 𝑏 < 1. One unit of capital costs $𝑘
(c) Use a graph of the plane 𝑧 = 𝑓 (𝑥, 𝑦) to explain and one unit of labor costs $𝓁. The price of the product
your answers in parts (a) and (b). is fixed at $𝑝 per unit.
34. Two products are manufactured in quantities 𝑞1 and 𝑞2 (a) If 𝑎 + 𝑏 < 1, how much capital and labor should
and sold at prices of 𝑝1 and 𝑝2 , respectively. The cost of the company use to maximize its profit?
producing them is given by (b) Is there a maximum profit in the case 𝑎 + 𝑏 = 1?
𝐶 = 2𝑞12 + 2𝑞22 + 10. What about 𝑎 + 𝑏 ≥ 1? Explain.

6 Adapted from M. Rosser and P. Lis, Basic Mathematics for Economists, 3rd ed. (New York: Routledge, 2016), p. 354.
824 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

38. Let 𝑓 (𝑥, 𝑦) = 𝑥2 (𝑦 + 1)3 + 𝑦2 . Show that 𝑓 has only How should 𝑝 be chosen to minimize the energy?
one critical point, namely (0, 0), and that point is a lo- (b) Now suppose the compression takes place in two
cal minimum but not a global minimum. Contrast this stages with two intermediate pressures, 𝑝1 and 𝑝2 .
with the case of a function with a single local minimum What choices of 𝑝1 and 𝑝2 minimize the energy if
in one-variable calculus.
( )2 ( )2 ( )2
39. Find the parabola of the form 𝑦 = 𝑎𝑥2 + 𝑏 which best 𝑝1 𝑝2 𝑃𝐹
𝐸= + + − 2?
fits the points (1, 0), (2, 2), (3, 4) by minimizing the sum 𝑃0 𝑝1 𝑝2
of squares, 𝑆, given by
𝑆 = (𝑎 + 𝑏)2 + (4𝑎 + 𝑏 − 2)2 + (9𝑎 + 𝑏 − 4)2 . 45. The Dorfman-Steiner rule shows how a company which
has a monopoly should set the price, 𝑝, of its product
40. For the data points (11, 16), (12, 17), (13, 17), and and how much advertising, 𝑎, it should buy. The price
(16, 20), find an expression for 𝑓 (𝑏, 𝑚), the sum of of advertising is 𝑝𝑎 per unit. The quantity, 𝑞, of the prod-
squared errors that are minimized on the least squares uct sold is given by 𝑞 = 𝐾𝑝−𝐸 𝑎𝜃 , where 𝐾 > 0, 𝐸 > 1,
line 𝑦 = 𝑏 + 𝑚𝑥. (You need not do the minimization.) and 0 < 𝜃 < 1 are constants. The cost to the company
41. Find the least squares line for the data points to make each item is 𝑐.
(0, 4), (1, 3), (2, 1).
(a) How does the quantity sold, 𝑞, change if the price,
42. Let 𝑓 (𝑥, 𝑦) = 80∕(𝑥𝑦) + 20𝑦 + 10𝑥 + 10𝑥𝑦 in the region 𝑝, increases? If the quantity of advertising, 𝑎, in-
𝑅 where 𝑥, 𝑦 > 0. creases?
(a) Explain why 𝑓 (𝑥, 𝑦) > 𝑓 (2, 1) at every point in 𝑅 (b) Show that the partial derivatives can be written in
where the form 𝜕𝑞∕𝜕𝑝 = −𝐸𝑞∕𝑝 and 𝜕𝑞∕𝜕𝑎 = 𝜃𝑞∕𝑎.
(i) 𝑥 > 20 (ii) 𝑦 > 20 (c) Explain why profit, 𝜋, is given by 𝜋 = 𝑝𝑞−𝑐𝑞−𝑝𝑎𝑎.
(d) If the company wants to maximize profit, what
(iii) 𝑥 < 0.01 and 𝑦 ≤ 20
must be true of the partial derivatives, 𝜕𝜋∕𝜕𝑝 and
(iv) 𝑦 < 0.01 and 𝑥 ≤ 20 𝜕𝜋∕𝜕𝑎?
(b) Explain why 𝑓 must have a global minimum at a (e) Find 𝜕𝜋∕𝜕𝑝 and 𝜕𝜋∕𝜕𝑎.
critical point in 𝑅. (f) Use your answers to parts (d) and (e) to show that
(c) Explain why 𝑓 must have a global minimum in 𝑅 at maximum profit,
at the point (2, 1).
𝑝−𝑐 1 𝑝−𝑐 𝑎
43. Let 𝑓 (𝑥, 𝑦) = 2∕𝑥+3∕𝑦+4𝑥+5𝑦 in the region 𝑅 where = and = .
𝑥, 𝑦 > 0. 𝑝 𝐸 𝑝𝑎 𝜃𝑞

(a) Explain why 𝑓 must have a global minimum at (g) By dividing your answers in part (f), show that at
some point in 𝑅. maximum profit,
(b) Find the global minimum.
𝑝𝑎 𝑎 𝜃
44. (a) The energy, 𝐸, required to compress a gas from a = .
𝑝𝑞 𝐸
fixed initial pressure 𝑃0 to a fixed final pressure 𝑃𝐹
through an intermediate pressure 𝑝 is7 This is the Dorfman-Steiner rule, that the ratio of
( )2 ( )2
𝑝 𝑃𝐹 the advertising budget to revenue does not depend
𝐸= + − 1. on the price of advertising.
𝑃0 𝑝

Strengthen Your Understanding

In Problems 46–48, explain what is wrong with the state- 49. A continuous function 𝑓 (𝑥, 𝑦) that has no global maxi-
ment. mum and no global minimum on the 𝑥𝑦-plane.

46. A function having no critical points in a region 𝑅 can- 50. A function 𝑓 (𝑥, 𝑦) and a region 𝑅 such that the maxi-
not have a global maximum in the region. mum value of 𝑓 on 𝑅 is on the boundary of 𝑅.
47. No continuous function has a global minimum on an Are the statements in Problems 51–59 true or false? Give
unbounded region 𝑅. reasons for your answer.
48. If 𝑓 (𝑥, 𝑦) has a local maximum value of 1 at the origin,
then the global maximum is 1. 51. If 𝑃0 is a global maximum of 𝑓 , where 𝑓 is defined on
all of 2-space, then 𝑃0 is also a local maximum of 𝑓 .
In Problems 49–50, give an example of:
52. Every function has a global maximum.
7 Adapted from Aris Rutherford, Discrete Dynamic Programming, p. 35 (New York: Blaisdell, 1964).
 15.3 CONSTRAINED OPTIMIZATION: LAGRANGE MULTIPLIERS 825

53. The region consisting of all points (𝑥, 𝑦) satisfying 𝑥2 + 57. If 𝑃 and 𝑄 are two distinct points in 2-space, and 𝑓 has
𝑦2 < 1 is bounded. a global maximum at 𝑃 , then 𝑓 cannot have a global
54. The region consisting of all points (𝑥, 𝑦) satisfying 𝑥2 + maximum at 𝑄.
𝑦2 < 1 is closed.
58. The function 𝑓 (𝑥, 𝑦) = sin(1 + 𝑒𝑥𝑦 ) must have a global
55. The function 𝑓 (𝑥, 𝑦) = 𝑥2 + 𝑦2 has a global minimum minimum in the square region 0 ≤ 𝑥 ≤ 1, 0 ≤ 𝑦 ≤ 1.
on the region 𝑥2 + 𝑦2 < 1.
56. The function 𝑓 (𝑥, 𝑦) = 𝑥2 + 𝑦2 has a global maximum 59. If 𝑃0 is a global minimum of 𝑓 on a closed and bounded
on the region 𝑥2 + 𝑦2 < 1. region, then 𝑃0 need not be a critical point of 𝑓 .

15.3 CONSTRAINED OPTIMIZATION: LAGRANGE MULTIPLIERS

Many, perhaps most, real optimization problems are constrained by external circumstances. For ex-
ample, a city wanting to build a public transportation system that will serve the greatest possible
number of people has only a limited number of tax dollars it can spend on the project. In this sec-
tion, we see how to find an optimum value under such constraints.
In Section 15.2, we saw how to optimize a function 𝑓 (𝑥, 𝑦) on a region 𝑅. If the region 𝑅 is the
entire 𝑥𝑦-plane, we have unconstrained optimization; if the region 𝑅 is not the entire 𝑥𝑦-plane, that
is, if 𝑥 or 𝑦 is restricted in some way, then we have constrained optimization.

Graphical Approach: Maximizing Production Subject to a Budget Constraint

Suppose we want to maximize production under a budget constraint. Suppose production, 𝑓 , is a
function of two variables, 𝑥 and 𝑦, which are quantities of two raw materials, and that

𝑓 (𝑥, 𝑦) = 𝑥2∕3 𝑦1∕3 .

If 𝑥 and 𝑦 are purchased at prices of 𝑝1 and 𝑝2 thousands of dollars per unit, what is the maximum
production 𝑓 that can be obtained with a budget of 𝑐 thousand dollars?
To maximize 𝑓 without regard to the budget, we simply increase 𝑥 and 𝑦. However, the budget
constraint prevents us from increasing 𝑥 and 𝑦 beyond a certain point. Exactly how does the budget
constrain us? With prices of 𝑝1 and 𝑝2 , the amount spent on 𝑥 is 𝑝1 𝑥 and the amount spent on 𝑦 is
𝑝2 𝑦, so we must have
𝑔(𝑥, 𝑦) = 𝑝1 𝑥 + 𝑝2 𝑦 ≤ 𝑐,
where 𝑔(𝑥, 𝑦) is the total cost of the raw materials and 𝑐 is the budget in thousands of dollars.
Let’s look at the case when 𝑝1 = 𝑝2 = 1 and 𝑐 = 3.78. Then

𝑥 + 𝑦 ≤ 3.78.

Figure 15.26 shows some contours of 𝑓 and the budget constraint represented by the line 𝑥+𝑦 =
3.78. Any point on or below the line represents a pair of values of 𝑥 and 𝑦 that we can afford. A point
on the line completely exhausts the budget, while a point below the line represents values of 𝑥 and
𝑦 which can be bought without using up the budget. Any point above the line represents a pair of
values that we cannot afford.
To maximize 𝑓 , we find the point which lies on the level curve with the largest possible value of
𝑓 and which lies within the budget. The point must lie on the budget constraint because production
is maximized when we spend all the available money. Unless we are at a point where the budget
constraint is tangent to a contour of 𝑓 , we can increase 𝑓 by moving in some direction along the line
representing the budget constraint in Figure 15.26. For example, if we are on the line to the left of
the point of tangency, moving right on the constraint will increase 𝑓 ; if we are on the line to the right
of the point of tangency, moving left will increase 𝑓 . Thus, the maximum value of 𝑓 on the budget
constraint occurs at the point where the budget constraint is tangent to the contour 𝑓 = 2.
826 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA
𝑦
Level curves of production

✛ 𝑓 =1
✛ 𝑓 =2
✛𝑓 =3

Maximum production
Budget constraint ✲
𝑥 + 𝑦 = 3.78 ✠
𝑃
𝑥
Figure 15.26: Optimal point, 𝑃 , where budget constraint
is tangent to a level of production function

Analytical Solution: Lagrange Multipliers

Figure 15.26 suggests that maximum production is achieved at the point where the budget constraint
is tangent to a level curve of the production function. The method of Lagrange multipliers uses this
fact in algebraic form. Figure 15.27 shows that at the optimum point, 𝑃 , the gradient of 𝑓 and the
normal to the budget line 𝑔(𝑥, 𝑦) = 𝑥 + 𝑦 = 3.78 are parallel. Thus, at 𝑃 , grad 𝑓 and grad 𝑔 are
parallel, so for some scalar 𝜆, called the Lagrange multiplier,
grad 𝑓 = 𝜆 grad 𝑔.
Calculating the gradients, we find that
2 −1∕3 1∕3 ⃗ 1 2∕3 −2∕3 ⃗
( ) ( ) ( )
𝑥 𝑦 𝑖 + 𝑥 𝑦 𝑗 = 𝜆 ⃗𝑖 + 𝑗⃗ .
3 3
Equating components gives
2 −1∕3 1∕3 1 2∕3 −2∕3
𝑥 𝑦 =𝜆 and 𝑥 𝑦 = 𝜆.
3 3
Eliminating 𝜆 gives
2 −1∕3 1∕3 1 2∕3 −2∕3
𝑥 𝑦 = 𝑥 𝑦 , which leads to 2𝑦 = 𝑥.
3 3
Since the constraint 𝑥 + 𝑦 = 3.78 must be satisfied, we have 𝑥 = 2.52 and 𝑦 = 1.26. Then
𝑓 (2.52, 1.26) = (2.52)2∕3(1.26)1∕3 ≈ 2.
As before, we see that the maximum value of 𝑓 is approximately 2. Thus, to maximize production
on a budget of $3780, we should use 2.52 units of one raw material and 1.26 units of the other.
𝑦
Level curves of production

✛ 𝑓 =1
✛ 𝑓 =2
✛𝑓 =3

grad 𝑓 = 𝜆 grad 𝑔
Budget constraint ✲ ✒
𝑔(𝑥, 𝑦) = 3.78
𝑃
✒
𝑥
Figure 15.27: At the point, 𝑃 , of maximum production,
the vectors grad 𝑓 and grad 𝑔 are parallel
 15.3 CONSTRAINED OPTIMIZATION: LAGRANGE MULTIPLIERS 827

Lagrange Multipliers in General

Suppose we want to optimize an objective function 𝑓 (𝑥, 𝑦) subject to a constraint 𝑔(𝑥, 𝑦) = 𝑐. We
look for extrema among the points which satisfy the constraint. We make the following definition.

Suppose 𝑃0 is a point satisfying the constraint 𝑔(𝑥, 𝑦) = 𝑐.

• 𝑓 has a local maximum at 𝑃0 subject to the constraint if 𝑓 (𝑃0 ) ≥ 𝑓 (𝑃 ) for all points
𝑃 near 𝑃0 satisfying the constraint.
• 𝑓 has a global maximum at 𝑃0 subject to the constraint if 𝑓 (𝑃0 ) ≥ 𝑓 (𝑃 ) for all points
𝑃 satisfying the constraint.
Local and global minima are defined similarly.

As we saw in the production example, constrained extrema occur at points of tangency of con-
tours of 𝑓 and 𝑔; they can also occur at endpoints of constraints. At a point of tangency, grad 𝑓 is
perpendicular to the constraint and so parallel to grad 𝑔. At interior points on the constraint where
grad 𝑓 is not perpendicular to the constraint, the value of 𝑓 can be increased or decreased by moving
along the constraint. Therefore constrained extrema occur only at points where grad 𝑓 and grad 𝑔
are parallel or at endpoints of the constraint. (See Figure 15.28.) At points where the gradients are
parallel, provided grad 𝑔 ≠ 0⃗ , there is a constant 𝜆 such that grad 𝑓 = 𝜆 grad 𝑔.

Optimizing 𝒇 Subject to the Constraint 𝒈 = 𝒄:

If a smooth function, 𝑓 , has a maximum or minimum subject to a smooth constraint 𝑔 = 𝑐 at
a point 𝑃0 , then either 𝑃0 satisfies the equations

grad 𝑓 = 𝜆 grad 𝑔 and 𝑔 = 𝑐,

or 𝑃0 is an endpoint of the constraint, or grad 𝑔(𝑃0 ) = 0⃗ . To investigate whether 𝑃0 is a global

maximum or minimum, compare values of 𝑓 at the points satisfying these three conditions.
The number 𝜆 is called the Lagrange multiplier.

If the set of points satisfying the constraint is closed and bounded, such as a circle or line seg-
ment, then there must be a global maximum and minimum of 𝑓 subject to the constraint. If the
constraint is not closed and bounded, such as a line or hyperbola, then there may or may not be a
global maximum and minimum.

grad 𝑓 𝑓 =4
✗ grad 𝑔
✗
𝑃0 𝑓 =3

𝑔=𝑐 grad 𝑓 𝑓 =2
✗

✎ grad 𝑔 𝑓 =1
Figure 15.28: Maximum and minimum values
of 𝑓 (𝑥, 𝑦) on 𝑔(𝑥, 𝑦) = 𝑐 are at points where
grad 𝑓 is parallel to grad 𝑔
828 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

Example 1 Find the maximum and minimum values of 𝑥 + 𝑦 on the circle 𝑥2 + 𝑦2 = 4.

Solution The objective function is

𝑓 (𝑥, 𝑦) = 𝑥 + 𝑦,
and the constraint is
𝑔(𝑥, 𝑦) = 𝑥2 + 𝑦2 = 4.
Since grad 𝑓 = 𝑓𝑥 ⃗𝑖 + 𝑓𝑦 𝑗⃗ = ⃗𝑖 + 𝑗⃗ and grad 𝑔 = 𝑔𝑥 ⃗𝑖 + 𝑔𝑦 𝑗⃗ = 2𝑥⃗𝑖 + 2𝑦𝑗⃗ , the condition grad 𝑓 =
𝜆 grad 𝑔 gives
1 = 2𝜆𝑥 and 1 = 2𝜆𝑦,
so
𝑥 = 𝑦.
We also know that
𝑥2 + 𝑦2 = 4,
√ √
giving 𝑥 = 𝑦 = 2 or 𝑥 = 𝑦 = − 2. The constraint ⃗
√ (it’s a circle) and grad 𝑔 ≠ 0
√ √ has no endpoints
√
on the circle, so we compare
√values at ( 2, 2) and (− 2, − 2). √
√ of 𝑓 √ Since √ 𝑓 (𝑥, 𝑦) = 𝑥√+ 𝑦, the
maximum value of 𝑓 is 𝑓 ( 2, 2) = 2 2; the minimum value is 𝑓 (− 2, − 2) = −2 2. (See
Figure 15.29.)

𝑥2 + 𝑦2 = 4 𝑦

𝑓 = −1 ❘ √ √ 𝑓
Maximum
( 2, 2)
𝑓 = −2
√
𝑓 = −2 2 𝑥 √
𝑓 =2 2
𝑓 =2
√ √
(− 2, − 2) 𝑓 =1
Minimum 𝑓
𝑓 =0
Figure 15.29: Maximum and minimum values of 𝑓 (𝑥, 𝑦) = 𝑥 + 𝑦 on the circle
𝑥2 + 𝑦2 = 4 are at points where contours of 𝑓 are tangent to the circle

How to Distinguish Maxima from Minima

There is a second-derivative test8 for classifying the critical points of constrained optimization prob-
lems, but it is more complicated than the test in Section 15.1. However, a graph of the constraint and
some contours usually shows which points are maxima, which points are minima, and which are
neither.

Optimization with Inequality Constraints

The production problem that we looked at first was to maximize production 𝑓 (𝑥, 𝑦) subject to a
budget constraint
𝑔(𝑥, 𝑦) = 𝑝1 𝑥 + 𝑝2 𝑦 ≤ 𝑐.
8 See J. E. Marsden and A. J. Tromba, Vector Calculus, 6th ed. (New York: W.H. Freeman, 2011), p. 220..
 15.3 CONSTRAINED OPTIMIZATION: LAGRANGE MULTIPLIERS 829

Since the inputs are nonnegative, 𝑥 ≥ 0 and 𝑦 ≥ 0, we have three inequality constraints, which
restrict (𝑥, 𝑦) to a region of the plane rather than to a curve in the plane. In principle, we should first
check to see whether or not 𝑓 (𝑥, 𝑦) has any critical points in the interior:

𝑝1 𝑥 + 𝑝2 𝑦 < 𝑐, 𝑥>0 𝑦 > 0.

However, in the case of a budget constraint, we can see that the maximum of 𝑓 must occur when the
budget is exhausted, so we look for the maximum value of 𝑓 on the boundary line:

𝑝1 𝑥 + 𝑝2 𝑦 = 𝑐, 𝑥≥0 𝑦 ≥ 0.

Strategy for Optimizing 𝑓 (𝑥, 𝑦) Subject to the Constraint 𝑔(𝑥, 𝑦) ≤ 𝑐

• Find all points in the region 𝑔(𝑥, 𝑦) < 𝑐 where grad 𝑓 is zero or undefined.
• Use Lagrange multipliers to find the local extrema of 𝑓 on the boundary 𝑔(𝑥, 𝑦) = 𝑐.
• Evaluate 𝑓 at the points found in the previous two steps and compare the values.

From Section 15.2 we know that if 𝑓 is continuous on a closed and bounded region, 𝑅, then 𝑓
is guaranteed to attain its global maximum and minimum values on 𝑅.

Example 2 Find the maximum and minimum values of 𝑓 (𝑥, 𝑦) = (𝑥 − 1)2 + (𝑦 − 2)2 subject to the constraint
𝑥2 + 𝑦2 ≤ 45.

Solution First, we look for all critical points of 𝑓 in the interior of the region. Setting

𝑓𝑥 (𝑥, 𝑦) = 2(𝑥 − 1) = 0
𝑓𝑦 (𝑥, 𝑦) = 2(𝑦 − 2) = 0,

we find 𝑓 has exactly one critical point at 𝑥 = 1, 𝑦 = 2. Since 12 + 22 < 45, that critical point is in
the interior of the region.
Next, we find the local extrema of 𝑓 on the boundary curve 𝑥2 + 𝑦2 = 45. To do this, we use
Lagrange multipliers with constraint 𝑔(𝑥, 𝑦) = 𝑥2 + 𝑦2 = 45. Setting grad 𝑓 = 𝜆 grad 𝑔, we get

2(𝑥 − 1) = 𝜆 ⋅ 2𝑥,
2(𝑦 − 2) = 𝜆 ⋅ 2𝑦.

We can’t have 𝑥 = 0 since the first equation would become −2 = 0. Similarly, 𝑦 ≠ 0. So we can
solve each equation for 𝜆 by dividing by 𝑥 and 𝑦. Setting the expressions for 𝜆 equal gives

𝑥−1 𝑦−2
= ,
𝑥 𝑦
so
𝑦 = 2𝑥.
Combining this with the constraint 𝑥2 + 𝑦2 = 45, we get

5𝑥2 = 45,

so
𝑥 = ±3.
Since 𝑦 = 2𝑥, we have possible local extrema at 𝑥 = 3, 𝑦 = 6 and 𝑥 = −3, 𝑦 = −6.
830 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

We conclude that the only candidates for the maximum and minimum values of 𝑓 in the region
occur at (1, 2), (3, 6), and (−3, −6). Evaluating 𝑓 at these three points, we find

𝑓 (1, 2) = 0, 𝑓 (3, 6) = 20, 𝑓 (−3, −6) = 80.

Therefore, the minimum value of 𝑓 is 0 at (1, 2) and the maximum value is 80 at (−3, −6).

The Meaning of 𝝀
In the uses of Lagrange multipliers so far, we never found (or needed) the value of 𝜆. However, 𝜆
does have a practical interpretation. In the production example, we wanted to maximize
𝑓 (𝑥, 𝑦) = 𝑥2∕3 𝑦1∕3
subject to the constraint
𝑔(𝑥, 𝑦) = 𝑥 + 𝑦 = 3.78.
We solved the equations
2 −1∕3 1∕3
𝑥 𝑦 = 𝜆,
3
1 2∕3 −2∕3
𝑥 𝑦 = 𝜆,
3
𝑥 + 𝑦 = 3.78,
to get 𝑥 = 2.52, 𝑦 = 1.26 and 𝑓 (2.52, 1.26) ≈ 2. Continuing to find 𝜆 gives us
𝜆 ≈ 0.53.
Now we do another, apparently unrelated, calculation. Suppose our budget is increased by one, from
3.78 to 4.78, giving a new budget constraint of 𝑥 + 𝑦 = 4.78. Then the corresponding solution is at
𝑥 = 3.19 and 𝑦 = 1.59 and the new maximum value (instead of 𝑓 = 2) is
𝑓 = (3.19)2∕3(1.59)1∕3 ≈ 2.53.
Notice that the amount by which 𝑓 has increased is 0.53, the value of 𝜆. Thus, in this example, the
value of 𝜆 represents the extra production achieved by increasing the budget by one—in other words,
the extra “bang” you get for an extra “buck” of budget. In fact, this is true in general:
• The value of 𝜆 is approximately the increase in the optimum value of 𝑓 when the budget is
increased by 1 unit.
More precisely:
• The value of 𝜆 represents the rate of change of the optimum value of 𝑓 as the budget increases.
An Expression for 𝝀
To interpret 𝜆, we look at how the optimum value of the objective function 𝑓 changes as the value
𝑐 of the constraint function 𝑔 is varied. In general, the optimum point (𝑥0 , 𝑦0 ) depends on the con-
straint value 𝑐. So, provided 𝑥0 and 𝑦0 are differentiable functions of 𝑐, we can use the chain rule to
differentiate the optimum value 𝑓 (𝑥0 (𝑐), 𝑦0 (𝑐)) with respect to 𝑐:
𝑑𝑓 𝜕𝑓 𝑑𝑥0 𝜕𝑓 𝑑𝑦0
= + .
𝑑𝑐 𝜕𝑥 𝑑𝑐 𝜕𝑦 𝑑𝑐
At the optimum point (𝑥0 , 𝑦0 ), we have 𝑓𝑥 = 𝜆𝑔𝑥 and 𝑓𝑦 = 𝜆𝑔𝑦 , and therefore
( )
𝑑𝑓 𝜕𝑔 𝑑𝑥0 𝜕𝑔 𝑑𝑦0 𝑑𝑔
=𝜆 + =𝜆 .
𝑑𝑐 𝜕𝑥 𝑑𝑐 𝜕𝑦 𝑑𝑐 𝑑𝑐
But, as 𝑔(𝑥0 (𝑐), 𝑦0 (𝑐)) = 𝑐, we see that 𝑑𝑔∕𝑑𝑐 = 1, so 𝑑𝑓 ∕𝑑𝑐 = 𝜆. Thus, we have the following
interpretation of the Lagrange multiplier 𝜆:
 15.3 CONSTRAINED OPTIMIZATION: LAGRANGE MULTIPLIERS 831

The value of 𝜆 is the rate of change of the optimum value of 𝑓 as 𝑐 increases (where 𝑔(𝑥, 𝑦) =
𝑐). If the optimum value of 𝑓 is written as 𝑓 (𝑥0 (𝑐), 𝑦0 (𝑐)), then

𝑑
𝑓 (𝑥0 (𝑐), 𝑦0 (𝑐)) = 𝜆.
𝑑𝑐

Example 3 The quantity of goods produced according to the function 𝑓 (𝑥, 𝑦) = 𝑥2∕3 𝑦1∕3 is maximized subject
to the budget constraint 𝑥 + 𝑦 ≤ 3.78. The budget is increased to allow for a small increase in
production. What is the price of the product if the sale of the additional goods covers the budget
increase?
Solution We know that 𝜆 = 0.53, which tells us that 𝑑𝑓 ∕𝑑𝑐 = 0.53. The constraint corresponds to a budget of
$3.78 thousand. Therefore increasing the budget by $1000 increases production by about 0.53 units.
In order to make the increase in budget profitable, the extra goods produced must sell for more than
$1000. Thus, if 𝑝 is the price of each unit of the good, then 0.53𝑝 is the revenue from the extra 0.53
units sold. Thus, we need 0.53𝑝 ≥ 1000 so 𝑝 ≥ 1000∕0.53 = $1890.

The Lagrangian Function

Constrained optimization problems are frequently solved using a Lagrangian function, . For exam-
ple, to optimize 𝑓 (𝑥, 𝑦) subject to the constraint 𝑔(𝑥, 𝑦) = 𝑐, we use the Lagrangian function
(𝑥, 𝑦, 𝜆) = 𝑓 (𝑥, 𝑦) − 𝜆(𝑔(𝑥, 𝑦) − 𝑐).
To see how the function  is used, compute the partial derivatives of :
𝜕 𝜕𝑓 𝜕𝑔
= −𝜆 ,
𝜕𝑥 𝜕𝑥 𝜕𝑥
𝜕 𝜕𝑓 𝜕𝑔
= −𝜆 ,
𝜕𝑦 𝜕𝑦 𝜕𝑦
𝜕
= −(𝑔(𝑥, 𝑦) − 𝑐).
𝜕𝜆
Notice that if (𝑥0 , 𝑦0 ) is an extreme point of 𝑓 (𝑥, 𝑦) subject to the constraint 𝑔(𝑥, 𝑦) = 𝑐 and 𝜆0 is
the corresponding Lagrange multiplier, then at the point (𝑥0 , 𝑦0 , 𝜆0 ) we have
𝜕 𝜕 𝜕
= 0 and =0 and = 0.
𝜕𝑥 𝜕𝑦 𝜕𝜆
In other words, (𝑥0 , 𝑦0 , 𝜆0 ) is a critical point for the unconstrained Lagrangian function, (𝑥, 𝑦, 𝜆).
Thus, the Lagrangian converts a constrained optimization problem to an unconstrained problem.

Example 4 A company has a production function with three inputs 𝑥, 𝑦, and 𝑧 given by

𝑓 (𝑥, 𝑦, 𝑧) = 50𝑥2∕5 𝑦1∕5 𝑧1∕5 .

The total budget is $24,000 and the company can buy 𝑥, 𝑦, and 𝑧 at $80, $12, and $10 per unit,
respectively. What combination of inputs will maximize production?9

Solution We need to maximize the objective function

𝑓 (𝑥, 𝑦, 𝑧) = 50𝑥2∕5 𝑦1∕5 𝑧1∕5 ,

subject to the constraint

𝑔(𝑥, 𝑦, 𝑧) = 80𝑥 + 12𝑦 + 10𝑧 = 24,000.
9 Adapted from M. Rosser and P. Lis, Basic Mathematics for Economists, 3rd ed. (New York: Routledge, 2016), p. 360.
832 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

The method for functions of two variables works for functions of three variables, so we construct
the Lagrangian function

(𝑥, 𝑦, 𝑧, 𝜆) = 50𝑥2∕5 𝑦1∕5 𝑧1∕5 − 𝜆(80𝑥 + 12𝑦 + 10𝑧 − 24,000),

and solve the system of equations we get from grad  = 0⃗ :

𝜕
= 20𝑥−3∕5 𝑦1∕5 𝑧1∕5 − 80𝜆 = 0,
𝜕𝑥
𝜕
= 10𝑥2∕5 𝑦−4∕5 𝑧1∕5 − 12𝜆 = 0,
𝜕𝑦
𝜕
= 10𝑥2∕5 𝑦1∕5 𝑧−4∕5 − 10𝜆 = 0,
𝜕𝑧
𝜕
= −(80𝑥 + 12𝑦 + 10𝑧 − 24,000) = 0.
𝜕𝜆
Simplifying this system gives

1 −3∕5 1∕5 1∕5

𝜆= 𝑥 𝑦 𝑧 ,
4
5
𝜆 = 𝑥2∕5 𝑦−4∕5 𝑧1∕5 ,
6
𝜆 = 𝑥2∕5 𝑦1∕5 𝑧−4∕5 ,
80𝑥 + 12𝑦 + 10𝑧 = 24,000.

Eliminating 𝑧 from the first two equations gives 𝑥 = 0.3𝑦. Eliminating 𝑥 from the second and third
equations gives 𝑧 = 1.2𝑦. Substituting for 𝑥 and 𝑧 into 80𝑥 + 12𝑦 + 10𝑧 = 24,000 gives

80(0.3𝑦) + 12𝑦 + 10(1.2𝑦) = 24,000,

so 𝑦 = 500. Then 𝑥 = 150 and 𝑧 = 600, and 𝑓 (150, 500, 600) = 4,622 units.
The graph of the constraint, 80𝑥 + 12𝑦 + 10𝑧 = 24,000, is a plane. Since the inputs 𝑥, 𝑦, 𝑧 must
be nonnegative, the graph is a triangle in the first octant, with edges on the coordinate planes. On
the boundary of the triangle, one (or more) of the variables 𝑥, 𝑦, 𝑧 is zero, so the function 𝑓 is zero.
Thus production is maximized within the budget using 𝑥 = 150, 𝑦 = 500, and 𝑧 = 600.

Exercises and Problems for Section 15.3 Online Resource: Additional Problems for Section 15.3
EXERCISES
In Exercises 1–18, use Lagrange multipliers to find the max- 10. 𝑓 (𝑥, 𝑦, 𝑧) = 𝑥 + 3𝑦 + 5𝑧, 𝑥2 + 𝑦2 + 𝑧2 = 1
imum and minimum values of 𝑓 subject to the given con-
straint, if such values exist. 11. 𝑓 (𝑥, 𝑦, 𝑧) = 𝑥2 − 𝑦2 − 2𝑧, 𝑥2 + 𝑦2 = 𝑧

1. 𝑓 (𝑥, 𝑦) = 𝑥 + 𝑦, 𝑥2 + 𝑦2 = 1 12. 𝑓 (𝑥, 𝑦, 𝑧) = 𝑥𝑦𝑧, 𝑥2 + 𝑦2 + 4𝑧2 = 12

2 2
2. 𝑓 (𝑥, 𝑦) = 𝑥 + 3𝑦 + 2, 𝑥 + 𝑦 = 10 13. 𝑓 (𝑥, 𝑦) = 𝑥2 + 2𝑦2 , 𝑥2 + 𝑦2 ≤ 4
2 2 2 2
3. 𝑓 (𝑥, 𝑦) = (𝑥 − 1) + (𝑦 + 2) , 𝑥 +𝑦 =5
3 2 2 14. 𝑓 (𝑥, 𝑦) = 𝑥 + 3𝑦, 𝑥2 + 𝑦2 ≤ 2
4. 𝑓 (𝑥, 𝑦) = 𝑥 + 𝑦, 3𝑥 + 𝑦 = 4
5. 𝑓 (𝑥, 𝑦) = 3𝑥 − 2𝑦, 𝑥2 + 2𝑦2 = 44 15. 𝑓 (𝑥, 𝑦) = 𝑥𝑦, 𝑥2 + 2𝑦2 ≤ 1
6. 𝑓 (𝑥, 𝑦) = 𝑥𝑦, 4𝑥2 + 𝑦2 = 8 16. 𝑓 (𝑥, 𝑦) = 𝑥3 + 𝑦, 𝑥+𝑦 ≥ 1
7. 𝑓 (𝑥, 𝑦) = 2𝑥𝑦, 5𝑥 + 4𝑦 = 100
17. 𝑓 (𝑥, 𝑦) = (𝑥 + 3)2 + (𝑦 − 3)2 , 𝑥2 + 𝑦2 ≤ 2
8. 𝑓 (𝑥1 , 𝑥2 ) = 𝑥1 2 + 𝑥2 2 , 𝑥1 + 𝑥2 = 1
9. 𝑓 (𝑥, 𝑦) = 𝑥2 + 𝑦, 𝑥2 − 𝑦2 = 1 18. 𝑓 (𝑥, 𝑦) = 𝑥2 𝑦 + 3𝑦2 − 𝑦, 𝑥2 + 𝑦2 ≤ 10
 15.3 CONSTRAINED OPTIMIZATION: LAGRANGE MULTIPLIERS 833

19. For each point marked in Figure 15.30, decide whether: In Exercises 20–23, a Cobb-Douglas production function
(a) The point is a local minimum, maximum, or nei- 𝑃 (𝐾, 𝐿) and budget 𝐵(𝐾, 𝐿) are given, where 𝐾 represents
ther for the function 𝑓 constrained by the loop. capital and 𝐿 represents labor. Use Lagrange multipliers to
(b) The point is a global minimum, maximum, or nei- find the values of 𝐾 and 𝐿 that maximize production given
ther subject to the constraint. a budget constraint or minimize budget given a production
constraint. Then give the value for 𝜆 and its meaning.
𝑓 = 10
𝑄
20. Maximize production: 𝑃 = 𝐾 1∕4 𝐿3∕4
Constraint Budget constraint: 𝐵 = 2𝐾 + 𝐿 = 40
𝑅
❘
𝑃 21. Maximize production: 𝑃 = 𝐾 2∕3 𝐿1∕3
Budget constraint: 𝐵 = 10𝐾 + 4𝐿 = 60
𝑓 = 20
𝑆
22. Maximize production: 𝑃 = 𝐾 2∕5 𝐿3∕5
𝑓 = 50 Budget constraint: 𝐵 = 4𝐾 + 5𝐿 = 100
𝑓 = 30 𝑓 = 40 𝑓 = 60
23. Minimize budget: 𝐵 = 4𝐾 + 𝐿
Figure 15.30 Production constraint: 𝑃 = 𝐾 1∕2 𝐿1∕2 = 200
PROBLEMS

24. Find the maximum value of 𝑓 (𝑥, 𝑦) = 𝑥 + 𝑦 − (𝑥 − 𝑦)2 29. Each person tries to balance his or her time between
on the triangular region 𝑥 ≥ 0, 𝑦 ≥ 0, 𝑥 + 𝑦 ≤ 1. leisure and work. The tradeoff is that as you work less
25. For 𝑓 (𝑥, 𝑦) = 𝑥2 + 6𝑥𝑦, find the global maximum your income falls. Therefore each person has indiffer-
and minimum on the closed region in the first quadrant ence curves which connect the number of hours of
bounded by the line 𝑥 + 𝑦 = 4 and the curve 𝑥𝑦 = 3. leisure, 𝑙, and income, 𝑠. If, for example, you are in-
different between 0 hours of leisure and an income of
26. (a) Draw contours of 𝑓 (𝑥, 𝑦) = 2𝑥 + 𝑦 for $1125 a week on the one hand, and 10 hours of leisure
𝑧 = −7, −5, −3, −1, 1, 3, 5, 7. and an income of $750 a week on the other hand, then
(b) On the same axes, graph the constraint 𝑥2 +𝑦2 = 5. the points 𝑙 = 0, 𝑠 = 1125, and 𝑙 = 10, 𝑠 = 750 both
(c) Use the graph to approximate the points at which lie on the same indifference curve. Table 15.3 gives in-
𝑓 has a maximum or a minimum value subject to formation on three indifference curves, I, II, and III.
the constraint 𝑥2 + 𝑦2 = 5.
(d) Use Lagrange multipliers to find the maximum and
minimum values of 𝑓 (𝑥, 𝑦) = 2𝑥 + 𝑦 subject to Table 15.3
𝑥2 + 𝑦2 = 5.
Weekly income Weekly leisure hours
27. Let 𝑓 (𝑥, 𝑦) = 𝑥𝛼 𝑦1−𝛼 for 0 < 𝛼 < 1. Find the value I II III I II III
of 𝛼 such that the maximum value of 𝑓 on the line
2𝑥 + 3𝑦 = 6 occurs at (1.5, 1). 1125 1250 1375 0 20 40
750 875 1000 10 30 50
28. Figure 15.31 shows contours of 𝑓 . Does 𝑓 have a max-
imum value subject to the constraint 𝑔(𝑥, 𝑦) = 𝑐 for 500 625 750 20 40 60
𝑥 ≥ 0, 𝑦 ≥ 0? If so, approximately where is it and what 375 500 625 30 50 70
is its value? Does 𝑓 have a minimum value subject to 250 375 500 50 70 90
the
𝑦 constraint? If so, approximately where and what?

𝑔(𝑥, 𝑦) = 𝑐 (a) Graph the three indifference curves.

16
(b) You have 100 hours a week available for work and
14
leisure combined, and you earn $10/hour. Write an
12 equation in terms of 𝑙 and 𝑠 which represents this
✠ 𝑓 = 100 𝑓 = 700
10 constraint.
8 𝑓 = 600 (c) On the same axes, graph this constraint.
(d) Estimate from the graph what combination of
6 𝑓 = 500 leisure hours and income you would choose under
4 ☛ 𝑓 = 400 these circumstances. Give the corresponding num-
✠ ✢
✠ 𝑓 = 300 ber of hours per week you would work.
2 ❂
✙ 𝑓 = 200
✾𝑥
2 4 6 8 10 12 14 16

Figure 15.31
 834 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

30. Figure 15.32 shows ∇𝑓 for a function 𝑓 (𝑥, 𝑦) and two 33. If the right side of the constraint in Exercise 5 is
curves 𝑔(𝑥, 𝑦) = 1 and 𝑔(𝑥, 𝑦) = 2. Mark the following: changed by the small amount Δ𝑐, by approximately how
much do the maximum and minimum values change?
(a) The point(s) 𝐴 where 𝑓 has a local maximum.
(b) The point(s) 𝐵 where 𝑓 has a saddle point. 34. If the right side of the constraint in Exercise 6 is
(c) The point 𝐶 where 𝑓 has a maximum on 𝑔 = 1. changed by the small amount Δ𝑐, by approximately how
(d) The point 𝐷 where 𝑓 has a minimum on 𝑔 = 1. much do the maximum and minimum values change?
(e) If you used Lagrange multipliers to find 𝐶, what 35. The function 𝑃 (𝑥, 𝑦) gives the number of units pro-
would the sign of 𝜆 be? Why? duced and 𝐶(𝑥, 𝑦) gives the cost of production.
(a) A company wishes to maximize production at a
fixed cost of $50,000. What is the objective func-
tion 𝑓 ? What is the constraint equation? What is
𝑔=2 ✲ the meaning of 𝜆 in this situation?
(b) A company wishes to minimize costs at a fixed pro-
duction level of 2000 units. What is the objective
𝑔=1 ✲ function 𝑓 ? What is the constraint equation? What
is the meaning of 𝜆 in this situation?

36. Design a closed cylindrical container which holds

100 cm3 and has the minimal possible surface area.
What should its dimensions be?
37. A company manufactures 𝑥 units of one item and 𝑦 units
Figure 15.32 of another. The total cost in dollars, 𝐶, of producing
31. The point 𝑃 is a maximum or minimum of the function these two items is approximated by the function
𝑓 subject to the constraint 𝑔(𝑥, 𝑦) = 𝑥 + 𝑦 = 𝑐, with
𝐶 = 5𝑥2 + 2𝑥𝑦 + 3𝑦2 + 800.
𝑥, 𝑦 ≥ 0. For the graphs (a) and (b), does 𝑃 give a max-
imum or a minimum of 𝑓 ? What is the sign of 𝜆? If 𝑃 (a) If the production quota for the total number of
gives a maximum, where does the minimum of 𝑓 oc- items (both types combined) is 39, find the min-
cur? If 𝑃 gives a minimum, where does the maximum imum production cost.
of 𝑓 occur? (b) Estimate the additional production cost or savings
(a) 𝑦 (b) 𝑦
if the production quota is raised to 40 or lowered
✛ 𝑓 =3 ✛ 𝑓 =1
to 38.
✛ 𝑓 =2 ✛ 𝑓 =2
✛𝑓 =1 ✛𝑓 =3 38. An international organization must decide how to spend
the $2,000,000 they have been allotted for famine relief
in a remote area. They expect to divide the money be-
✛ 𝑔(𝑥, 𝑦) = 𝑐 ✛ 𝑔(𝑥, 𝑦) = 𝑐
tween buying rice at $38.5/sack and beans at $35/sack.
The number, 𝑃 , of people who would be fed if they buy
𝑃 𝑥 𝑃 𝑥 𝑥 sacks of rice and 𝑦 sacks of beans is given by
32. Figure 15.33 shows the optimal point (marked with a 𝑥𝑦
𝑃 = 1.1𝑥 + 𝑦 − .
dot) in three optimization problems with the same con- 108
straint. Arrange the corresponding values of 𝜆 in in-
What is the maximum number of people that can be fed,
creasing order. (Assume 𝜆 is positive.)
(I) 𝑦 (II) 𝑦 and how should the organization allocate its money?
39. The quantity, 𝑞, of a product manufactured depends on
𝑓 =3 the number of workers, 𝑊 , and the amount of capital
𝑓 =2 invested, 𝐾, and is given by
𝑓 =3
𝑓 =2 ❄
✠ 𝑞 = 6𝑊 3∕4 𝐾 1∕4 .
𝑓 =1 ✛ 𝑓 =1
𝑥 𝑥
𝑦 Labor costs are $10 per worker and capital costs are $20
(III)
per unit, and the budget is $3000.
(a) What are the optimum number of workers and the
𝑓 =3 optimum number of units of capital?
𝑓 =2
(b) Show that at the optimum values of 𝑊 and 𝐾,
❄
✠
✛ the ratio of the marginal productivity of labor
𝑓 =1
𝑥 (𝜕𝑞∕𝜕𝑊 ) to the marginal productivity of capital
Figure 15.33
 15.3 CONSTRAINED OPTIMIZATION: LAGRANGE MULTIPLIERS 835

(𝜕𝑞∕𝜕𝐾) is the same as the ratio of the cost of a 𝐿, labor

10,
unit of labor to the cost of a unit of capital. 20 𝐵 = 000
(c) Recompute the optimum values of 𝑊 and 𝐾 when 800
0 𝑃 = 90
the budget is increased by one dollar. Check that in- 15 60
00 80
creasing the budget by $1 allows the production of
𝜆 extra units of the good, where 𝜆 is the Lagrange 70
multiplier. 10 40 60
00
50
40. A neighborhood health clinic has a budget of $600,000
5 40
per quarter. The director of the clinic wants to allocate 30
the budget to maximize the number of patient visits, 𝑉 , 200 20
which is given as a function of the number of doctors, 0 𝐾 , capital
5 10 15 20
𝐷, and the number of nurses, 𝑁, by
𝑉 = 1000𝐷0.6 𝑁 0.3 . Figure 15.34

A doctor gets $40,000 per quarter; nurses get $10,000

per quarter.
(a) Set up the director’s constrained optimization 43. A doctor wants to schedule visits for two patients who
problem. have been operated on for tumors so as to minimize the
(b) Describe, in words, the conditions which must be expected delay in detecting a new tumor. Visits for pa-
satisfied by 𝜕𝑉 ∕𝜕𝐷 and 𝜕𝑉 ∕𝜕𝑁 for 𝑉 to have an tients 1 and 2 are scheduled at intervals of 𝑥1 and 𝑥2
optimum value. weeks, respectively. A total of 𝑚 visits per week is avail-
(c) Solve the problem formulated in part (a). able for both patients combined.
(d) Find the value of the Lagrange multiplier and in- The recurrence rates for tumors for patients 1 and
terpret its meaning in this problem. 2 are judged to be 𝑣1 and 𝑣2 tumors per week, respec-
(e) At the optimum point, what is the marginal cost tively. Thus, 𝑣1 ∕(𝑣1 + 𝑣2 ) and 𝑣2 ∕(𝑣1 + 𝑣2 ) are the prob-
of a patient visit (that is, the cost of an additional abilities that patient 1 and patient 2, respectively, will
visit)? Will that marginal cost rise or fall with the have the next tumor. It is known that the expected de-
number of visits? Why? lay in detecting a tumor for a patient checked every 𝑥
weeks is 𝑥∕2. Hence, the expected detection delay for
41. (a) In Problem 39, does the value of 𝜆 change if the
both patients combined is given by10
budget changes from $3000 to $4000?
(b) In Problem 40, does the value of 𝜆 change if the 𝑣1 𝑥 𝑣2 𝑥
budget changes from $600,000 to $700,000? 𝑓 (𝑥1 , 𝑥2 ) = ⋅ 1 + ⋅ 2.
𝑣1 + 𝑣2 2 𝑣1 + 𝑣2 2
(c) What condition must a Cobb-Douglas production
function, 𝑄 = 𝑐𝐾 𝑎 𝐿𝑏 , satisfy to ensure that the Find the values of 𝑥1 and 𝑥2 in terms of 𝑣1 and 𝑣2 that
marginal increase of production (that is, the rate of minimize 𝑓 (𝑥1 , 𝑥2 ) subject to the fact that 𝑚, the num-
increase of production with budget) is not affected ber of visits per week, is fixed.
by the size of the budget?
42. The production function 𝑃 (𝐾, 𝐿) gives the number of 44. What is the value of the Lagrange multiplier in Prob-
pairs of skis produced per week at a factory operating lem 43? What are the units of 𝜆? What is its practical
with 𝐾 units of capital and 𝐿 units of labor. The con- significance to the doctor?
tour diagram for 𝑃 is in Figure 15.34; the parallel lines 45. Figure 15.35 shows two weightless springs with spring
are budget constraints for budgets, 𝐵, in dollars. constants 𝑘1 and 𝑘2 attached between a ceiling and floor
(a) On each budget constraint, mark the point that without tension or compression. A mass 𝑚 is placed be-
gives the maximum production. tween the springs which settle into equilibrium as in
(b) Complete the table, where the budget, 𝐵, is in dol- Figure 15.36. The magnitudes 𝑓1 and 𝑓2 of the forces
lars and the maximum production is the number of of the springs on the mass minimize the complementary
pairs of skis to be produced each week. energy
𝑓12 𝑓2
𝐵 2000 4000 6000 8000 10000 + 2
2𝑘1 2𝑘2
𝑀
subject to the force balance constraint 𝑓1 + 𝑓2 = 𝑚𝑔.
(c) Estimate the Lagrange multiplier 𝜆 = 𝑑𝑀∕𝑑𝐵 at (a) Determine 𝑓1 and 𝑓2 by the method of Lagrange
a budget of $6000. Give units for the multiplier. multipliers.
10 Adapted from Daniel Kent, Ross Shachter, et al., “Efficient Scheduling of Cystoscopies in Monitoring for Recurrent

Bladder Cancer,” Medical Decision Making (Philadelphia: Hanley and Belfus, 1989).
836 Chapter 15 OPTIMIZATION: LOCAL AND GLOBAL EXTREMA

(b) If you are familiar with Hooke’s law, find the mean- 𝑉 (𝑝, 𝑞, 𝐼) is the maximum of 𝑈 (𝑥, 𝑦) subject to the con-
ing of 𝜆. straint 𝑝𝑥 + 𝑞𝑦 = 𝐼.
(a) What is the practical meaning of 𝑉 (𝑝, 𝑞, 𝐼)?
(b) The Lagrange multiplier 𝜆 that arises in the max-
imization defining 𝑉 is called the marginal utility
𝑘1 of money. What is its practical meaning?
(c) Find formulas for 𝑉 (𝑝, 𝑞, 𝐼) and 𝜆 if 𝑈 (𝑥, 𝑦) = 𝑥𝑦.

51. The function ℎ(𝑥, 𝑦) = 𝑥2 + 𝑦2 − 𝜆(2𝑥 + 4𝑦 − 15) has a

minimum value 𝑚(𝜆) for each value of 𝜆.
m
𝑘2
(a) Find 𝑚(𝜆).
(b) For which value of 𝜆 is 𝑚(𝜆) the largest and what
Figure 15.35 Figure 15.36 is that maximum value?
(c) Find the minimum value of 𝑓 (𝑥, 𝑦) = 𝑥2 + 𝑦2 sub-
ject to the constraint 2𝑥+4𝑦 = 15 using the method
∑3
46. (a) If 𝑖=1 𝑥𝑖 = 1, find the values of 𝑥1 , 𝑥2 , 𝑥3 making of Lagrange multipliers and evaluate 𝜆.
∑3 (d) Compare your answers to parts (b) and (c).
𝑥 2 minimum.
𝑖=1 𝑖
(b) Generalize the result of part (a) to find the mini- 52. Let 𝑓 be differentiable and grad 𝑓 (2, 1) = −3⃗𝑖 + 4𝑗⃗ .
∑𝑛 ∑𝑛
mum value of 𝑖=1 𝑥𝑖 2 subject to 𝑖=1 𝑥𝑖 = 1. You want to see if (2, 1) is a candidate for the maxi-
47. Let 𝑓 (𝑥, 𝑦) = 𝑎𝑥2 + 𝑏𝑥𝑦 + 𝑐𝑦2 . Show that the maximum mum and minimum values of 𝑓 subject to a constraint
value of 𝑓 (𝑥, 𝑦) subject to the constraint 𝑥2 + 𝑦2 = 1 is satisfied by the point (2, 1).
equal to 𝜆, the Lagrange multiplier.
(a) Show (2, 1) is not a candidate if the constraint is
48. Find the minimum distance from the point (1, 2, 10) to 𝑥2 + 𝑦2 = 5.
the paraboloid given by the equation 𝑧 = 𝑥2 + 𝑦2 . Give (b) Show (2, 1) is a candidate if the constraint is (𝑥 −
a geometric justification for your answer. 5)2 + (𝑦 + 3)2 = 25. From a sketch of the contours
49. A company produces one product from two inputs (for for 𝑓 near (2, 1) and the constraint, decide whether
example, capital and labor). Its production function (2, 1) is a candidate for a maximum or minimum.
𝑔(𝑥, 𝑦) gives the quantity of the product that can be pro- (c) Do the same as part (b), but using the constraint
duced with 𝑥 units of the first input and 𝑦 units of the (𝑥 + 1)2 + (𝑦 − 5)2 = 25.
second. The cost function (or expenditure function) is
53. A person’s satisfaction from consuming a quantity 𝑥1
the three-variable function 𝐶(𝑝, 𝑞, 𝑢) where 𝑝 and 𝑞 are
of one item and a quantity 𝑥2 of another item is given
the unit prices of the two inputs. For fixed 𝑝, 𝑞, and 𝑢,
by
the value 𝐶(𝑝, 𝑞, 𝑢) is the minimum of 𝑓 (𝑥, 𝑦) = 𝑝𝑥+𝑞𝑦
subject to the constraint 𝑔(𝑥, 𝑦) = 𝑢. 𝑆 = 𝑢(𝑥1 , 𝑥2 ) = 𝑎 ln 𝑥1 + (1 − 𝑎) ln 𝑥2 ,

(a) What is the practical meaning of 𝐶(𝑝, 𝑞, 𝑢)? where 𝑎 is a constant, 0 < 𝑎 < 1. The prices of the two
(b) Find a formula for 𝐶(𝑝, 𝑞, 𝑢) if 𝑔(𝑥, 𝑦) = 𝑥𝑦. items are 𝑝1 and 𝑝2 respectively, and the budget is 𝑏.
(a) Express the maximum satisfaction that can be
50. A utility function 𝑈 (𝑥, 𝑦) for two items gives the util-
achieved as a function of 𝑝1 , 𝑝2 , and 𝑏.
ity (benefit) to a consumer of 𝑥 units of item 1 and 𝑦
(b) Find the amount of money that must be spent to
units of item 2. The indirect utility function is the three-
achieve a particular level of satisfaction, 𝑐, as a
variable function 𝑉 (𝑝, 𝑞, 𝐼) where 𝑝 and 𝑞 are the unit
function of 𝑝1 , 𝑝2 , and 𝑐.
prices of the two items. For fixed 𝑝, 𝑞, and 𝐼, the value

Strengthen Your Understanding

In Problems 54–55, explain what is wrong with the state- In Problems 56–60, give an example of:
ment.
56. A function 𝑓 (𝑥, 𝑦) whose maximum subject to the con-
straint 𝑥2 + 𝑦2 = 5 is at (3, 4).
54. The function 𝑓 (𝑥, 𝑦) = 𝑥𝑦 has a maximum of 2 on the
constraint 𝑥 + 𝑦 = 2. 57. A function 𝑓 (𝑥, 𝑦) to be optimized with constraint 𝑥2 +
2𝑦2 ≤ 1 such that the minimum value does not change
55. If the level curves of 𝑓 (𝑥, 𝑦) and the level curves of when the constraint is changed to 𝑥2 + 𝑦2 ≤ 1 + 𝑐 for
𝑔(𝑥, 𝑦) are not tangent at any point on the constraint 𝑐 > 0.
𝑔(𝑥, 𝑦) = 𝑐, 𝑥 ≥ 0, 𝑦 ≥ 0, then 𝑓 has no maximum 58. A function 𝑓 (𝑥, 𝑦) with a minimum at (1, 1) on the con-
on the constraint. straint 𝑥 + 𝑦 = 2.
 15.3 CONSTRAINED OPTIMIZATION: LAGRANGE MULTIPLIERS 837

59. A function 𝑓 (𝑥, 𝑦) that has a maximum but no minimum 64. If 𝑓 (𝑥, 𝑦) has a local maximum at (𝑎, 𝑏) subject to the
on the constraint 𝑥 + 𝑦 = 4. constraint 𝑔(𝑥, 𝑦) = 𝑐, then grad𝑓 (𝑎, 𝑏) = 0⃗ .
60. A contour diagram of a function 𝑓 whose maximum 65. The function 𝑓 (𝑥, 𝑦) = 𝑥 + 𝑦 has no global maximum
value on the constraint 𝑥 + 2𝑦 = 6, 𝑥 ≥ 0, 𝑦 ≥ 0 occurs subject to the constraint 𝑥 − 𝑦 = 0.
at one of the endpoints.
66. The point (2, −1) is a local minimum of 𝑓 (𝑥, 𝑦) =
For Problems 61–62, use Figure 15.37. The grid lines are 𝑥2 + 𝑦2 subject to the constraint 𝑥 + 2𝑦 = 0.
one unit apart. 67. If grad 𝑓 (𝑎, 𝑏) and grad 𝑔(𝑎, 𝑏) point in opposite direc-
𝑦 tions, then (𝑎, 𝑏) is a local minimum of 𝑓 (𝑥, 𝑦) con-
strained by 𝑔(𝑥, 𝑦) = 𝑐.

In Problems 68–75, suppose that 𝑀 and 𝑚 are the maxi-

𝑔=𝑐
mum and minimum values of 𝑓 (𝑥, 𝑦) subject to the con-
𝑓=
𝑓=
5 straint 𝑔(𝑥, 𝑦) = 𝑐 and that (𝑎, 𝑏) satisfies 𝑔(𝑎, 𝑏) = 𝑐. Decide
4 whether the statements are true or false. Give an explanation
𝑓=
3 for your answer.
𝑓=
2
68. If 𝑓 (𝑎, 𝑏) = 𝑀, then 𝑓𝑥 (𝑎, 𝑏) = 𝑓𝑦 (𝑎, 𝑏) = 0.
𝑓=
1
69. If 𝑓 (𝑎, 𝑏) = 𝑀, then 𝑓 (𝑎, 𝑏) = 𝜆𝑔(𝑎, 𝑏) for some value
of 𝜆.

𝑓 =0
70. If grad 𝑓 (𝑎, 𝑏) = 𝜆 grad 𝑔(𝑎, 𝑏), then 𝑓 (𝑎, 𝑏) = 𝑀 or
𝑥 𝑓 (𝑎, 𝑏) = 𝑚.
Figure 15.37 71. If 𝑓 (𝑎, 𝑏) = 𝑀 and 𝑓𝑥 (𝑎, 𝑏)∕𝑓𝑦 (𝑎, 𝑏) = 5, then
𝑔𝑥 (𝑎, 𝑏)∕𝑔𝑦 (𝑎, 𝑏) = 5.
61. Find the maximum and minimum values of 𝑓 on 𝑔 = 𝑐. 72. If 𝑓 (𝑎, 𝑏) = 𝑚 and 𝑔𝑥 (𝑎, 𝑏) = 0, then 𝑓𝑥 (𝑎, 𝑏) = 0.
At which points do they occur?
73. Increasing the value of 𝑐 increases the value of 𝑀.
62. Find the maximum and minimum values of 𝑓 on the
triangular region below 𝑔 = 𝑐 in the first quadrant. 74. Suppose that 𝑓 (𝑎, 𝑏) = 𝑀 and that grad 𝑓 (𝑎, 𝑏) =
3 grad 𝑔(𝑎, 𝑏). Then increasing the value of 𝑐 by 0.02
Are the statements in Problems 63–67 true or false? Give increases the value of 𝑀 by about 0.06.
reasons for your answer. 75. Suppose that 𝑓 (𝑎, 𝑏) = 𝑚 and that grad 𝑓 (𝑎, 𝑏) =
63. If 𝑓 (𝑥, 𝑦) has a local maximum at (𝑎, 𝑏) subject to the 3 grad 𝑔(𝑎, 𝑏). Then increasing the value of 𝑐 by 0.02
constraint 𝑔(𝑥, 𝑦) = 𝑐, then 𝑔(𝑎, 𝑏) = 𝑐. decreases the value of 𝑚 by about 0.06.

Online Resource: Review problems and Projects